id
string | submitter
string | authors
string | title
string | comments
string | journal-ref
string | doi
string | report-no
string | categories
string | license
string | abstract
string | versions
list | update_date
timestamp[s] | authors_parsed
sequence |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
2506.05159 | Zhi-Meng Li | Zhi-Meng Li, Yong Zhang | Search for binarity in Asymptotic Giant Branch stars utilizing the future Chinese Space Station Telescope (CSST) | 16 pages, 4 figures, submitted to RAA | null | null | null | astro-ph.SR astro-ph.IM | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Binary systems in the Asymptotic Giant Branch (AGB) phase are widely recognized as a leading theoretical framework underpinning the observed asymmetric morphologies of planetary nebulae. However, the detection of binary companions in AGB systems is severely hampered by the overwhelming brightness and variability of the evolved primary star, which dominate the photo-metric and spectroscopic signatures. Ultraviolet (UV) excess emission has been proposed as a candidate diagnostic for the presence of binary companions in AGB systems. This paper evaluates the Chinese Space Station Telescope's (CSST) ability to detect UV excess emission in AGB stars, leveraging its unprecedented UV sensitivity and wide-field survey capabilities. We employed synthetic spectral libraries of M0-M8 type giants for primary stars and the ATLAS 9 atmospheric model grid for companion stars spanning a temperature range of 6500 K to 12000 K. By convolving these model spectra with the CSST multi-band filter system, we computed color-color diagrams (g-y versus NUV-u) to construct a diagnostic grid. This grid incorporates interstellar extinction corrections and establishes a framework for identifying AGB binary candidates through direct comparison between observed photometry and theoretical predictions. Furthermore, we discuss the physical origins of UV excess in AGB stars. This study pioneers a diagnostic framework leveraging CSST's unique multi-band UV-visible synergy to construct color-color grids for binary candidate identification, overcoming limitations of non-simultaneous multi-instrument observations. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 15:38:01 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Li",
"Zhi-Meng",
""
],
[
"Zhang",
"Yong",
""
]
] |
2506.05160 | Guilherme Fiusa | Guilherme Fiusa, Pedro E. Harunari, Abhaya S. Hegde, Gabriel T. Landi | A framework for fluctuating times and counting observables in stochastic excursions | Companion paper to arXiv:2505.06208 | null | null | null | cond-mat.stat-mech quant-ph | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Many natural systems exhibit dynamics characterized by alternating phases. Describing the fluctuations of such systems over stochastic trajectories is necessary across diverse fields, from biological motors to quantum thermal machines. In an accompanying Letter, we introduced the notion of stochastic excursions -- a framework to analyze far from equilibrium processes via sub-trajectories. Through counting observables, this framework captures finite-time fluctuations and trajectory-level behavior, which provides insights into thermodynamical trade-offs between thermodynamic quantities of interest, such as entropy production and dynamical activity. In this work, we enhance this formalism by providing a suite of technical results on how to efficiently compute excursion-related quantities. Our analytical results provide explicit formulas for general moments of counting variables and excursion duration, as well as their covariance and conditional moments. We show that excursion statistics recover full counting statistics results, and uncover a relation between fluctuations of counting observables at single-excursion level and the steady state diffusion coefficient (noise). We also discuss a fluctuation theorem for individual excursions. In addition, we explore how analyzing excursions and using the results developed here can yield insights into three problems of interest: the three-qubit absorption refrigerator, cellular sensing, and birth-and-death processes. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 15:38:18 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Fiusa",
"Guilherme",
""
],
[
"Harunari",
"Pedro E.",
""
],
[
"Hegde",
"Abhaya S.",
""
],
[
"Landi",
"Gabriel T.",
""
]
] |
2506.05161 | Benjamin Rose | B. M. Rose, M. Vincenzi, R. Hounsell, H. Qu, L. Aldoroty, D. Scolnic, R. Kessler, P. Macias, D. Brout, M. Acevedo, R. C. Chen, S. Gomez, E. Peterson, D. Rubin, M. Sako | The Hourglass Simulation: A Catalog for the Roman High-Latitude Time-Domain Core Community Survey | submitted to ApJ | null | null | null | astro-ph.IM astro-ph.CO astro-ph.HE | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | We present a simulation of the time-domain catalog for the Nancy Grace Roman Space Telescope's High-Latitude Time-Domain Core Community Survey. This simulation, called the Hourglass simulation, uses the most up-to-date spectral energy distribution models and rate measurements for ten extra-galactic time-domain sources. We simulate these models through the design reference Roman Space Telescope survey: four filters per tier, a five day cadence, over two years, a wide tier of 19 deg$^2$ and a deep tier of 4.2 deg$^2$, with $\sim$20% of those areas also covered with prism observations. We find that a science-independent Roman time-domain catalog, assuming a S/N at max of >5, would have approximately 21,000 Type Ia supernovae, 40,000 core-collapse supernovae, around 70 superluminous supernovae, $\sim$35 tidal disruption events, 3 kilonovae, and possibly pair-instability supernovae. In total, Hourglass has over 64,000 transient objects, 11 million photometric observations, and 500,000 spectra. Additionally, Hourglass is a useful data set to train machine learning classification algorithms. We show that SCONE is able to photometrically classify Type Ia supernovae with high precision ($\sim$95%) to a z > 2. Finally, we present the first realistic simulations of non-Type Ia supernovae spectral-time series data from Roman's prism. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 15:38:33 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Rose",
"B. M.",
""
],
[
"Vincenzi",
"M.",
""
],
[
"Hounsell",
"R.",
""
],
[
"Qu",
"H.",
""
],
[
"Aldoroty",
"L.",
""
],
[
"Scolnic",
"D.",
""
],
[
"Kessler",
"R.",
""
],
[
"Macias",
"P.",
""
],
[
"Brout",
"D.",
""
],
[
"Acevedo",
"M.",
""
],
[
"Chen",
"R. C.",
""
],
[
"Gomez",
"S.",
""
],
[
"Peterson",
"E.",
""
],
[
"Rubin",
"D.",
""
],
[
"Sako",
"M.",
""
]
] |
2506.05162 | Christian Copetti | Davide Bason, Christian Copetti, Lorenzo Di Pietro and Ziming Ji | $\mathcal{N}=2$ Super-Yang Mills in AdS$_4$ and $F_{\text{AdS}}$-maximization | 29 pages plus (several) appendices, comments are welcome | null | null | null | hep-th | http://creativecommons.org/licenses/by/4.0/ | We investigate the dynamics of four-dimensional $\mathcal{N}=2$ $SU(2)$ super Yang--Mills theory on an AdS background. We propose that the boundary conditions that preserve the AdS super-isometries are determined by maximizing the real part of the AdS partition function $F_{\text{AdS}}=-\log Z_{\text{AdS}}$. At weak coupling $\Lambda L \ll 1$ the maximization singles out the Dirichlet boundary condition with an $SU(2)$ boundary global symmetry, corresponding to the classical vacuum at the origin of the Coulomb branch with fully un-higgsed gauge group. We find that for $\Lambda L \sim \mathcal{O}(1)$ new boundary conditions are favored, with gauge-group higgsed down to $U(1)$, matching the expectation from the flat space limit. We use supersymmetric localization to compute $Z_{\text{AdS}}$ nonperturbatively. We further provide evidence for a relation between $F_{\text{AdS}}$ and the $\mathcal{N}=2$ prepotential in AdS background. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 15:38:38 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Bason",
"Davide",
""
],
[
"Copetti",
"Christian",
""
],
[
"Di Pietro",
"Lorenzo",
""
],
[
"Ji",
"Ziming",
""
]
] |
2506.05163 | Gabriele Magrini | Gabriele Magrini, Niccol\`o Marini, Federico Becattini, Lorenzo Berlincioni, Niccol\`o Biondi, Pietro Pala, Alberto Del Bimbo | FRED: The Florence RGB-Event Drone Dataset | null | null | null | null | cs.CV | http://creativecommons.org/licenses/by/4.0/ | Small, fast, and lightweight drones present significant challenges for traditional RGB cameras due to their limitations in capturing fast-moving objects, especially under challenging lighting conditions. Event cameras offer an ideal solution, providing high temporal definition and dynamic range, yet existing benchmarks often lack fine temporal resolution or drone-specific motion patterns, hindering progress in these areas. This paper introduces the Florence RGB-Event Drone dataset (FRED), a novel multimodal dataset specifically designed for drone detection, tracking, and trajectory forecasting, combining RGB video and event streams. FRED features more than 7 hours of densely annotated drone trajectories, using 5 different drone models and including challenging scenarios such as rain and adverse lighting conditions. We provide detailed evaluation protocols and standard metrics for each task, facilitating reproducible benchmarking. The authors hope FRED will advance research in high-speed drone perception and multimodal spatiotemporal understanding. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 15:40:41 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Magrini",
"Gabriele",
""
],
[
"Marini",
"Niccolò",
""
],
[
"Becattini",
"Federico",
""
],
[
"Berlincioni",
"Lorenzo",
""
],
[
"Biondi",
"Niccolò",
""
],
[
"Pala",
"Pietro",
""
],
[
"Del Bimbo",
"Alberto",
""
]
] |
2506.05164 | Andrea Loretti | Pasquale Cascarano, Andrea Loretti, Matteo Martinoni, Luca Zanuttini, Alessio Di Pasquale and Gustavo Marfia | Towards the target and not beyond: 2d vs 3d visual aids in mr-based neurosurgical simulation | 15 pages, 7 figures, 3 tables, journal | null | null | null | cs.GR | http://creativecommons.org/licenses/by/4.0/ | Neurosurgery increasingly uses Mixed Reality (MR) technologies for intraoperative assistance. The greatest challenge in this area is mentally reconstructing complex 3D anatomical structures from 2D slices with millimetric precision, which is required in procedures like External Ventricular Drain (EVD) placement. MR technologies have shown great potential in improving surgical performance, however, their limited availability in clinical settings underscores the need for training systems that foster skill retention in unaided conditions. In this paper, we introduce NeuroMix, an MR-based simulator for EVD placement. We conduct a study with 48 participants to assess the impact of 2D and 3D visual aids on usability, cognitive load, technology acceptance, and procedure precision and execution time. Three training modalities are compared: one without visual aids, one with 2D aids only, and one combining both 2D and 3D aids. The training phase takes place entirely on digital objects, followed by a freehand EVD placement testing phase performed with a physical catherer and a physical phantom without MR aids. We then compare the participants performance with that of a control group that does not undergo training. Our findings show that participants trained with both 2D and 3D aids achieve a 44\% improvement in precision during unaided testing compared to the control group, substantially higher than the improvement observed in the other groups. All three training modalities receive high usability and technology acceptance ratings, with significant equivalence across groups. The combination of 2D and 3D visual aids does not significantly increase cognitive workload, though it leads to longer operation times during freehand testing compared to the control group. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 15:41:04 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Cascarano",
"Pasquale",
""
],
[
"Loretti",
"Andrea",
""
],
[
"Martinoni",
"Matteo",
""
],
[
"Zanuttini",
"Luca",
""
],
[
"Di Pasquale",
"Alessio",
""
],
[
"Marfia",
"Gustavo",
""
]
] |
2506.05165 | Suhan Park | Dongwoo Son, Suhan Park | LiPo: A Lightweight Post-optimization Framework for Smoothing Action Chunks Generated by Learned Policies | 6 pages, 7 figures, 1 table | null | null | null | cs.RO | http://creativecommons.org/licenses/by/4.0/ | Recent advances in imitation learning have enabled robots to perform increasingly complex manipulation tasks in unstructured environments. However, most learned policies rely on discrete action chunking, which introduces discontinuities at chunk boundaries. These discontinuities degrade motion quality and are particularly problematic in dynamic tasks such as throwing or lifting heavy objects, where smooth trajectories are critical for momentum transfer and system stability. In this work, we present a lightweight post-optimization framework for smoothing chunked action sequences. Our method combines three key components: (1) inference-aware chunk scheduling to proactively generate overlapping chunks and avoid pauses from inference delays; (2) linear blending in the overlap region to reduce abrupt transitions; and (3) jerk-minimizing trajectory optimization constrained within a bounded perturbation space. The proposed method was validated on a position-controlled robotic arm performing dynamic manipulation tasks. Experimental results demonstrate that our approach significantly reduces vibration and motion jitter, leading to smoother execution and improved mechanical robustness. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 15:41:23 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Son",
"Dongwoo",
""
],
[
"Park",
"Suhan",
""
]
] |
2506.05166 | Procheta Sen | Bhavik Chandna, Zubair Bashir, Procheta Sen | Dissecting Bias in LLMs: A Mechanistic Interpretability Perspective | null | null | null | null | cs.CL cs.AI | http://creativecommons.org/licenses/by/4.0/ | Large Language Models (LLMs) are known to exhibit social, demographic, and gender biases, often as a consequence of the data on which they are trained. In this work, we adopt a mechanistic interpretability approach to analyze how such biases are structurally represented within models such as GPT-2 and Llama2. Focusing on demographic and gender biases, we explore different metrics to identify the internal edges responsible for biased behavior. We then assess the stability, localization, and generalizability of these components across dataset and linguistic variations. Through systematic ablations, we demonstrate that bias-related computations are highly localized, often concentrated in a small subset of layers. Moreover, the identified components change across fine-tuning settings, including those unrelated to bias. Finally, we show that removing these components not only reduces biased outputs but also affects other NLP tasks, such as named entity recognition and linguistic acceptability judgment because of the sharing of important components with these tasks. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 15:43:34 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Chandna",
"Bhavik",
""
],
[
"Bashir",
"Zubair",
""
],
[
"Sen",
"Procheta",
""
]
] |
2506.05167 | Yeonseok Jeong | Yeonseok Jeong, Jinsu Kim, Dohyeon Lee, Seung-won Hwang | ECoRAG: Evidentiality-guided Compression for Long Context RAG | null | null | null | null | cs.CL cs.AI cs.IR | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Large Language Models (LLMs) have shown remarkable performance in Open-Domain Question Answering (ODQA) by leveraging external documents through Retrieval-Augmented Generation (RAG). To reduce RAG overhead, from longer context, context compression is necessary. However, prior compression methods do not focus on filtering out non-evidential information, which limit the performance in LLM-based RAG. We thus propose Evidentiality-guided RAG, or \textbf{ECoRAG} framework. ECoRAG improves LLM performance by compressing retrieved documents based on evidentiality, ensuring whether answer generation is supported by the correct evidence. As an additional step, ECoRAG reflects whether the compressed content provides sufficient evidence, and if not, retrieves more until sufficient. Experiments show that ECoRAG improves LLM performance on ODQA tasks, outperforming existing compression methods. Furthermore, ECoRAG is highly cost-efficient, as it not only reduces latency but also minimizes token usage by retaining only the necessary information to generate the correct answer. Code is available at https://github.com/ldilab/ECoRAG. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 15:43:49 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Jeong",
"Yeonseok",
""
],
[
"Kim",
"Jinsu",
""
],
[
"Lee",
"Dohyeon",
""
],
[
"Hwang",
"Seung-won",
""
]
] |
2506.05168 | Yunsheng Tian | Yunsheng Tian, Joshua Jacob, Yijiang Huang, Jialiang Zhao, Edward Gu, Pingchuan Ma, Annan Zhang, Farhad Javid, Branden Romero, Sachin Chitta, Shinjiro Sueda, Hui Li, Wojciech Matusik | Fabrica: Dual-Arm Assembly of General Multi-Part Objects via Integrated Planning and Learning | null | null | null | null | cs.RO | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Multi-part assembly poses significant challenges for robots to execute long-horizon, contact-rich manipulation with generalization across complex geometries. We present Fabrica, a dual-arm robotic system capable of end-to-end planning and control for autonomous assembly of general multi-part objects. For planning over long horizons, we develop hierarchies of precedence, sequence, grasp, and motion planning with automated fixture generation, enabling general multi-step assembly on any dual-arm robots. The planner is made efficient through a parallelizable design and is optimized for downstream control stability. For contact-rich assembly steps, we propose a lightweight reinforcement learning framework that trains generalist policies across object geometries, assembly directions, and grasp poses, guided by equivariance and residual actions obtained from the plan. These policies transfer zero-shot to the real world and achieve 80% successful steps. For systematic evaluation, we propose a benchmark suite of multi-part assemblies resembling industrial and daily objects across diverse categories and geometries. By integrating efficient global planning and robust local control, we showcase the first system to achieve complete and generalizable real-world multi-part assembly without domain knowledge or human demonstrations. Project website: http://fabrica.csail.mit.edu/ | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 15:43:52 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Tian",
"Yunsheng",
""
],
[
"Jacob",
"Joshua",
""
],
[
"Huang",
"Yijiang",
""
],
[
"Zhao",
"Jialiang",
""
],
[
"Gu",
"Edward",
""
],
[
"Ma",
"Pingchuan",
""
],
[
"Zhang",
"Annan",
""
],
[
"Javid",
"Farhad",
""
],
[
"Romero",
"Branden",
""
],
[
"Chitta",
"Sachin",
""
],
[
"Sueda",
"Shinjiro",
""
],
[
"Li",
"Hui",
""
],
[
"Matusik",
"Wojciech",
""
]
] |
2506.05169 | Weicheng Gao | Weicheng Gao | Through-the-Wall Radar Human Activity Recognition WITHOUT Using Neural Networks | 15 pages, 8 figures, 8 tables | null | null | null | cs.CV eess.SP | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | After a few years of research in the field of through-the-wall radar (TWR) human activity recognition (HAR), I found that we seem to be stuck in the mindset of training on radar image data through neural network models. The earliest related works in this field based on template matching did not require a training process, and I believe they have never died. Because these methods possess a strong physical interpretability and are closer to the basis of theoretical signal processing research. In this paper, I would like to try to return to the original path by attempting to eschew neural networks to achieve the TWR HAR task and challenge to achieve intelligent recognition as neural network models. In detail, the range-time map and Doppler-time map of TWR are first generated. Then, the initial regions of the human target foreground and noise background on the maps are determined using corner detection method, and the micro-Doppler signature is segmented using the multiphase active contour model. The micro-Doppler segmentation feature is discretized into a two-dimensional point cloud. Finally, the topological similarity between the resulting point cloud and the point clouds of the template data is calculated using Mapper algorithm to obtain the recognition results. The effectiveness of the proposed method is demonstrated by numerical simulated and measured experiments. The open-source code of this work is released at: https://github.com/JoeyBGOfficial/Through-the-Wall-Radar-Human-Activity-Recognition-Without-Using-Neural-Networks. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 15:45:08 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Gao",
"Weicheng",
""
]
] |
2506.05170 | Dario Panici | Dario Panici, Eduardo Rodriguez, Rory Conlin, Daniel Dudt, Egemen Kolemen | Extending near-axis equilibria in DESC | 30 pages, 7 figures | null | null | null | physics.plasm-ph | http://creativecommons.org/licenses/by/4.0/ | The near-axis description of optimised stellarator fields has proven to be a powerful tool both for design and understanding of this magnetic confinement concept. The description consists of an asymptotic model of the equilibrium in the distance from its centermost axis, and is thus only approximate. Any practical application therefore requires the eventual construction of a global equilibrium. This paper presents a novel way of constructing global equilibria using the \texttt{DESC} code that guarantees the correct asymptotic behaviour imposed by a given near-axis construction. The theoretical underpinnings of this construction are carefully presented, and benchmarking examples provided. This opens the door to an efficient coupling of the near-axis framework and that of global equilibria for future optimisation efforts. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 15:45:29 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Panici",
"Dario",
""
],
[
"Rodriguez",
"Eduardo",
""
],
[
"Conlin",
"Rory",
""
],
[
"Dudt",
"Daniel",
""
],
[
"Kolemen",
"Egemen",
""
]
] |
2506.05171 | Linxuan He | Linxuan He, Qing-Shan Jia, Ang Li, Hongyan Sang, Ling Wang, Jiwen Lu, Tao Zhang, Jie Zhou, Yi Zhang, Yisen Wang, Peng Wei, Zhongyuan Wang, Henry X. Liu, and Shuo Feng | Towards provable probabilistic safety for scalable embodied AI systems | null | null | null | null | eess.SY cs.SY | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Embodied AI systems, comprising AI models and physical plants, are increasingly prevalent across various applications. Due to the rarity of system failures, ensuring their safety in complex operating environments remains a major challenge, which severely hinders their large-scale deployment in safety-critical domains, such as autonomous vehicles, medical devices, and robotics. While achieving provable deterministic safety--verifying system safety across all possible scenarios--remains theoretically ideal, the rarity and complexity of corner cases make this approach impractical for scalable embodied AI systems. To address this challenge, we introduce provable probabilistic safety, which aims to ensure that the residual risk of large-scale deployment remains below a predefined threshold. Instead of attempting exhaustive safety proof across all corner cases, this paradigm establishes a probabilistic safety boundary on overall system performance, leveraging statistical methods to enhance feasibility and scalability. A well-defined probabilistic safety boundary enables embodied AI systems to be deployed at scale while allowing for continuous refinement of safety guarantees. Our work focuses on three core questions: what is provable probabilistic safety, how to prove the probabilistic safety, and how to achieve the provable probabilistic safety. By bridging the gap between theoretical safety assurance and practical deployment, our work offers a pathway toward safer, large-scale adoption of embodied AI systems in safety-critical applications. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 15:46:25 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"He",
"Linxuan",
""
],
[
"Jia",
"Qing-Shan",
""
],
[
"Li",
"Ang",
""
],
[
"Sang",
"Hongyan",
""
],
[
"Wang",
"Ling",
""
],
[
"Lu",
"Jiwen",
""
],
[
"Zhang",
"Tao",
""
],
[
"Zhou",
"Jie",
""
],
[
"Zhang",
"Yi",
""
],
[
"Wang",
"Yisen",
""
],
[
"Wei",
"Peng",
""
],
[
"Wang",
"Zhongyuan",
""
],
[
"Liu",
"Henry X.",
""
],
[
"Feng",
"Shuo",
""
]
] |
2506.05172 | Weichen Shi | Weichen Shi | A Framework for Ethical Judgment of Smart City Applications | null | null | null | null | cs.CY | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | As modern cities increasingly adopt a variety of sensors and Internet of Things (IoT) technologies to collect and analyze data about residents, environments, and public services, they are fostering greater interactions among smart city applications, residents, governments, and businesses. This trend makes it essential for regulators to focus on these interactions to manage smart city practices effectively and prevent unethical outcomes. To facilitate ethical analysis for smart city applications, this paper introduces a judgment framework that examines various scenarios where ethical issues may arise. Employing a multi-agent approach, the framework incorporates diverse social entities and applies logic-based ethical rules to identify potential violations. Through a rights-based analysis, we developed a set of 13 ethical principles and rules to guide ethical practices in smart cities. We utilized two specification languages, Prototype Verification System (PVS) and Alloy, to model our multi-agent system. Our analysis suggests that Alloy may be more efficient for formalizing smart cities and conducting ethical rule checks, particularly with the assistance of a human evaluator. Simulations of a real-world smart city application demonstrate that our ethical judgment framework effectively detects unethical outcomes and can be extended for practical use. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 15:47:43 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Shi",
"Weichen",
""
]
] |
2506.05173 | Guglielmo Lockhart | Guglielmo Lockhart and Luca Novelli | The hyperplane string, RCFTs, and the swampland | 44+1 pages, 2 figures | null | null | null | hep-th | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Six dimensional $\mathcal{N}=(1,0)$ supergravity features BPS strings whose properties encode highly nontrivial information about the parent 6d theory. We focus on a distinguished set of theories whose string charge lattice is one-dimensional. In geometric theories, the generator of the lattice arises from a D3 brane wrapping the hyperplane class in $\mathbb{P}^2$. This hyperplane string is expected to remain stable even when one ventures beyond the geometric regime where it becomes challenging to verify which candidate 6d theories belong to the swampland. We identify five 6d models which from the perspective of the hyperplane string deviate the most from being geometric. For these theories we are able to provide an exact description of the left-moving sector of the hyperplane string worldsheet in terms of a rational conformal field theory and provide evidence for their consistency. In one instance, using RCFT methods we are able to determine the elliptic genus and find that in the unflavored limit it matches with the elliptic genus of geometric models. We argue that the non-geometric model is connected to geometric ones via a sequence of Higgsing transitions. These results lead us to formulate a proposal relating the quantum corrected moduli space of the hyperplane string CFT with a region of the landscape of 6d $(1,0)$ quantum gravity. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 15:47:55 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Lockhart",
"Guglielmo",
""
],
[
"Novelli",
"Luca",
""
]
] |
2506.05174 | Yifan Zhang | Yifan Zhang, Joe Kileel | Norming Sets for Tensor and Polynomial Sketching | 16 pages | null | null | null | math.NA cs.IT cs.NA math.AG math.IT | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | This paper develops the sketching (i.e., randomized dimension reduction) theory for real algebraic varieties and images of polynomial maps, including, e.g., the set of low rank tensors and tensor networks. Through the lens of norming sets, we provide a framework for controlling the sketching dimension for \textit{any} sketch operator used to embed said sets, including sub-Gaussian, fast Johnson-Lindenstrauss, and tensor structured sketch operators. Leveraging norming set theory, we propose a new sketching method called the median sketch. It embeds such a set $V$ using only $\widetilde{\mathcal{O}}(\dim V)$ tensor structured or sparse linear measurements. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 15:49:23 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Zhang",
"Yifan",
""
],
[
"Kileel",
"Joe",
""
]
] |
2506.05175 | Yuzhi Huang | Yuzhi Huang, Chenxin Li, Haitao Zhang, Zixu Lin, Yunlong Lin, Hengyu Liu, Wuyang Li, Xinyu Liu, Jiechao Gao, Yue Huang, Xinghao Ding, Yixuan Yuan | Track Any Anomalous Object: A Granular Video Anomaly Detection Pipeline | null | null | null | null | cs.CV | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Video anomaly detection (VAD) is crucial in scenarios such as surveillance and autonomous driving, where timely detection of unexpected activities is essential. Although existing methods have primarily focused on detecting anomalous objects in videos -- either by identifying anomalous frames or objects -- they often neglect finer-grained analysis, such as anomalous pixels, which limits their ability to capture a broader range of anomalies. To address this challenge, we propose a new framework called Track Any Anomalous Object (TAO), which introduces a granular video anomaly detection pipeline that, for the first time, integrates the detection of multiple fine-grained anomalous objects into a unified framework. Unlike methods that assign anomaly scores to every pixel, our approach transforms the problem into pixel-level tracking of anomalous objects. By linking anomaly scores to downstream tasks such as segmentation and tracking, our method removes the need for threshold tuning and achieves more precise anomaly localization in long and complex video sequences. Experiments demonstrate that TAO sets new benchmarks in accuracy and robustness. Project page available online. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 15:49:39 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Huang",
"Yuzhi",
""
],
[
"Li",
"Chenxin",
""
],
[
"Zhang",
"Haitao",
""
],
[
"Lin",
"Zixu",
""
],
[
"Lin",
"Yunlong",
""
],
[
"Liu",
"Hengyu",
""
],
[
"Li",
"Wuyang",
""
],
[
"Liu",
"Xinyu",
""
],
[
"Gao",
"Jiechao",
""
],
[
"Huang",
"Yue",
""
],
[
"Ding",
"Xinghao",
""
],
[
"Yuan",
"Yixuan",
""
]
] |
2506.05176 | Mingxin Li | Yanzhao Zhang, Mingxin Li, Dingkun Long, Xin Zhang, Huan Lin, Baosong Yang, Pengjun Xie, An Yang, Dayiheng Liu, Junyang Lin, Fei Huang, Jingren Zhou | Qwen3 Embedding: Advancing Text Embedding and Reranking Through Foundation Models | null | null | null | null | cs.CL | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | In this work, we introduce the Qwen3 Embedding series, a significant advancement over its predecessor, the GTE-Qwen series, in text embedding and reranking capabilities, built upon the Qwen3 foundation models. Leveraging the Qwen3 LLMs' robust capabilities in multilingual text understanding and generation, our innovative multi-stage training pipeline combines large-scale unsupervised pre-training with supervised fine-tuning on high-quality datasets. Effective model merging strategies further ensure the robustness and adaptability of the Qwen3 Embedding series. During the training process, the Qwen3 LLMs serve not only as backbone models but also play a crucial role in synthesizing high-quality, rich, and diverse training data across multiple domains and languages, thus enhancing the training pipeline. The Qwen3 Embedding series offers a spectrum of model sizes (0.6B, 4B, 8B) for both embedding and reranking tasks, addressing diverse deployment scenarios where users can optimize for either efficiency or effectiveness. Empirical evaluations demonstrate that the Qwen3 Embedding series achieves state-of-the-art results across diverse benchmarks. Notably, it excels on the multilingual evaluation benchmark MTEB for text embedding, as well as in various retrieval tasks, including code retrieval, cross-lingual retrieval and multilingual retrieval. To facilitate reproducibility and promote community-driven research and development, the Qwen3 Embedding models are publicly available under the Apache 2.0 license. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 15:49:48 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Zhang",
"Yanzhao",
""
],
[
"Li",
"Mingxin",
""
],
[
"Long",
"Dingkun",
""
],
[
"Zhang",
"Xin",
""
],
[
"Lin",
"Huan",
""
],
[
"Yang",
"Baosong",
""
],
[
"Xie",
"Pengjun",
""
],
[
"Yang",
"An",
""
],
[
"Liu",
"Dayiheng",
""
],
[
"Lin",
"Junyang",
""
],
[
"Huang",
"Fei",
""
],
[
"Zhou",
"Jingren",
""
]
] |
2506.05177 | Darius K\"ohnke | D. K\"ohnke, T. Bayer, M. Wollenhaupt | Adiabatic dynamics in a V-type quantum system by oppositely chirped counterrotating circularly polarized laser pulses | 15 pages, 5 figures | null | null | null | physics.atom-ph quant-ph | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Shaped free electron vortices (SEVs) have recently been studied using atomic (1+2) resonance-enhanced multiphoton ionization (REMPI) by oppositely chirped counterrotating circularly polarized (OC-CRCP) femtosecond laser pulses. By transitioning from the perturbative to the non-perturbative REMPI regime, we identify an adiabatic excitation mechanism in a resonant V-type three-level system, termed V-RAP due to its similarities to rapid adiabatic passage (RAP). Experimentally, we observe a pronounced change in the shape of the measured three-dimensional photoelectron momentum distribution (3D PMD), which we trace back to this mechanism via analytical calculations and numerical simulations of the bound state and ionization dynamics. In V-RAP, the atom adiabatically follows the OC-CRCP field, with the two excited states driven in anti-phase, leading to an adiabatic cancellation of specific ionization pathways and explaining the observed changes in the PMD. In the experiment, we combine supercontinuum polarization pulse shaping to generate OC-CRCP femtosecond laser pulses with velocity map imaging-based photoelectron tomography to reconstruct the 3D PMD. The reconstructed PMDs are decomposed by 3D Fourier analysis into SEVs of different rotational symmetry, revealing a significant enhancement of the $c_6$-symmetric contribution, which is the signature of the V-RAP. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 15:50:32 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Köhnke",
"D.",
""
],
[
"Bayer",
"T.",
""
],
[
"Wollenhaupt",
"M.",
""
]
] |
2506.05178 | Joshua Hess | Joshua Hess and Quaid Morris | Associative Memory and Generative Diffusion in the Zero-noise Limit | null | null | null | null | cs.LG cond-mat.dis-nn math.DS nlin.AO q-bio.NC | http://creativecommons.org/licenses/by/4.0/ | Connections between generative diffusion and continuous-state associative memory models are studied. Morse-Smale dynamical systems are emphasized as universal approximators of gradient-based associative memory models and diffusion models as white-noise perturbed systems thereof. Universal properties of associative memory that follow from this description are described and used to characterize a generic transition from generation to memory as noise levels diminish. Structural stability inherited by Morse-Smale flows is shown to imply a notion of stability for diffusions at vanishing noise levels. Applied to one- and two-parameter families of gradients, this indicates stability at all but isolated points of associative memory learning landscapes and the learning and generation landscapes of diffusion models with gradient drift in the zero-noise limit, at which small sets of generic bifurcations characterize qualitative transitions between stable systems. Examples illustrating the characterization of these landscapes by sequences of these bifurcations are given, along with structural stability criterion for classic and modern Hopfield networks (equivalently, the attention mechanism). | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 15:51:47 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Hess",
"Joshua",
""
],
[
"Morris",
"Quaid",
""
]
] |
2506.05179 | Zolt\'an Guba | Zolt\'an Guba, Aris Alexandradinata, Tom\'a\v{s} Bzdu\v{s}ek | Delicate Wannier insulators | 20 pages (including 11 figures, 1 table, and bibliography) | null | null | null | cond-mat.mes-hall | http://creativecommons.org/licenses/by/4.0/ | The defining feature of topological insulators is that their valence states are not continuously deformable to a suitably defined atomic limit without breaking the symmetry or closing the energy gap. When the atomic limit is given by symmetric exponentially-localized Wannier orbitals, one finds stable and fragile topological insulators characterized by robust bulk-boundary correspondence. More recently, delicate topological insulators (DIs) have been introduced, whose metallic states are guaranteed only at sharply terminated edges and surfaces. Although Wannierizable, their Wannier orbitals necessarily span multiple unit cells, thus refining the notion of the atomic limit. In this work, we extend delicate topological invariants from Bloch states to hybrid Wannier functions. The resulting models, dubbed delicate Wannier insulators (DWIs), are deformable to unicellular atomic limit in the absence of edges and surfaces; nevertheless, they exhibit obstructions to such deformations as well as topological boundary states in the presence of sharply terminated hinges and corners. We present a layering construction that allows us to elevate a DI in $d$ dimensions into a DWI in $(d\,{+}\,1)$ dimensions. We illustrate the phenomenology of DWIs by deploying the layering construction on three concrete models. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 15:52:18 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Guba",
"Zoltán",
""
],
[
"Alexandradinata",
"Aris",
""
],
[
"Bzdušek",
"Tomáš",
""
]
] |
2506.05180 | Eliseo Pavone | Eliseo Pavone and Luigi Tedesco | Viscosity in Isotropic Cosmological Backgrounds in General Relativity and Starobinsky Gravity | null | null | null | null | gr-qc | http://creativecommons.org/licenses/by/4.0/ | We present a general analysis of the role of shear viscosity in cosmological backgrounds, focusing on isotropic space-time in both Einstein and $f(R)$ gravity. By computing the divergence of the stress-energy tensor in a general class of isotropic (but not necessarily homogeneous) geometries, we show that shear viscosity does not contribute to the background dynamics when the fluid is comoving. This result holds in both the Jordan and Einstein frames, and implies that shear viscosity cannot affect the electromagnetic luminosity distance which is determined by the background light-like geodesics. As an application of our results, we critically examine recent claims that shear viscosity can alter the Hubble evolution and the electromagnetic luminosity distance in Starobinsky gravity. We demonstrate that the continuity equation used in that work is at odds both with the covariant conservation of the stress-energy tensor and the local second law of thermodynamics. We further show that even in models where such modifications could mimic bulk viscosity, the resulting entropy evolution is inconsistent with standard thermodynamic expectations. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 15:52:40 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Pavone",
"Eliseo",
""
],
[
"Tedesco",
"Luigi",
""
]
] |
2506.05181 | En Li | En Li, Manish Kumar, Xinnan Peng, Tong Shen, Diego Soler-Polo, Yu Wang, Yu Teng, Haoyu Zhang, Shaotang Song, Jishan Wu, Pavel Jelinek, Jiong Lu | Designer polyradical nanographenes with strong spin entanglement and perturbation resilience via Clar's goblet extension | null | null | null | null | cond-mat.mes-hall cond-mat.mtrl-sci | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Polyradical nanographenes featuring strong spin entanglement and robust many-body spin states against external magnetic perturbations not only enable the exploration of correlated quantum magnetism at the molecular scale, but also constitute promising candidates for developing molecular qubits with chemical tunability and building scalable quantum networks. Here, we employed a predictive design strategy to achieve the on-surface synthesis of two homologues of Clar goblet, C62H22 and C76H26, via lateral and vertical extensions of the parent structure, respectively. Vertical extension increases the number of topologically frustrated zero-energy modes, which scale linearly with the total number of benzene ring rows. In contrast, the lateral extension enhances electron-electron interactions, leading to the emergence of additional radical states beyond those predicted by the topological zero-energy modes. Consequently, both structures exhibit correlated tetraradical character and a many-body singlet ground state as confirmed by multireference theoretical calculations. These magnetic states arise from unique magnetic origins and also display distinct resilience to external perturbations, which can be experimentally validated using nickelocene-functionalized scanning probe techniques. Our work presents a general strategy for rational design of highly entangled polyradical nanographenes with tunable spin numbers and resilience of their many-body spin states to perturbations, opening exciting possibilities for exploring novel correlated spin phases in molecular systems and advancing quantum information technologies. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 15:52:42 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Li",
"En",
""
],
[
"Kumar",
"Manish",
""
],
[
"Peng",
"Xinnan",
""
],
[
"Shen",
"Tong",
""
],
[
"Soler-Polo",
"Diego",
""
],
[
"Wang",
"Yu",
""
],
[
"Teng",
"Yu",
""
],
[
"Zhang",
"Haoyu",
""
],
[
"Song",
"Shaotang",
""
],
[
"Wu",
"Jishan",
""
],
[
"Jelinek",
"Pavel",
""
],
[
"Lu",
"Jiong",
""
]
] |
2506.05182 | Shivani Jayantkumar Upadhyay | Shivani Upadhyay, Messiah Ataey, Shariyar Murtuza, Yifan Nie, Jimmy Lin | On the Comprehensibility of Multi-structured Financial Documents using LLMs and Pre-processing Tools | 15 pages, 5 figures, 9 tables | null | null | null | cs.IR | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | The proliferation of complex structured data in hybrid sources, such as PDF documents and web pages, presents unique challenges for current Large Language Models (LLMs) and Multi-modal Large Language Models (MLLMs) in providing accurate answers. Despite the recent advancements of MLLMs, they still often falter when interpreting intricately structured information, such as nested tables and multi-dimensional plots, leading to hallucinations and erroneous outputs. This paper explores the capabilities of LLMs and MLLMs in understanding and answering questions from complex data structures found in PDF documents by leveraging industrial and open-source tools as part of a pre-processing pipeline. Our findings indicate that GPT-4o, a popular MLLM, achieves an accuracy of 56% on multi-structured documents when fed documents directly, and that integrating pre-processing tools raises the accuracy of LLMs to 61.3% for GPT-4o and 76% for GPT-4, and with lower overall cost. The code is publicly available at https://github.com/OGCDS/FinancialQA. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 15:52:44 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Upadhyay",
"Shivani",
""
],
[
"Ataey",
"Messiah",
""
],
[
"Murtuza",
"Shariyar",
""
],
[
"Nie",
"Yifan",
""
],
[
"Lin",
"Jimmy",
""
]
] |
2506.05183 | Zhicheng Yang | Zhicheng Yang, Zhijiang Guo, Yinya Huang, Xiaodan Liang, Yiwei Wang, Jing Tang | TreeRPO: Tree Relative Policy Optimization | 13pages, 6 figures | null | null | null | cs.LG cs.AI | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Large Language Models (LLMs) have shown remarkable reasoning capabilities through Reinforcement Learning with Verifiable Rewards (RLVR) methods. However, a key limitation of existing approaches is that rewards defined at the full trajectory level provide insufficient guidance for optimizing the intermediate steps of a reasoning process. To address this, we introduce \textbf{\name}, a novel method that estimates the mathematical expectations of rewards at various reasoning steps using tree sampling. Unlike prior methods that rely on a separate step reward model, \name directly estimates these rewards through this sampling process. Building on the group-relative reward training mechanism of GRPO, \name innovatively computes rewards based on step-level groups generated during tree sampling. This advancement allows \name to produce fine-grained and dense reward signals, significantly enhancing the learning process and overall performance of LLMs. Experimental results demonstrate that our \name algorithm substantially improves the average Pass@1 accuracy of Qwen-2.5-Math on test benchmarks, increasing it from 19.0\% to 35.5\%. Furthermore, \name significantly outperforms GRPO by 2.9\% in performance while simultaneously reducing the average response length by 18.1\%, showcasing its effectiveness and efficiency. Our code will be available at \href{https://github.com/yangzhch6/TreeRPO}{https://github.com/yangzhch6/TreeRPO}. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 15:56:38 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Yang",
"Zhicheng",
""
],
[
"Guo",
"Zhijiang",
""
],
[
"Huang",
"Yinya",
""
],
[
"Liang",
"Xiaodan",
""
],
[
"Wang",
"Yiwei",
""
],
[
"Tang",
"Jing",
""
]
] |
2506.05184 | Neeraj Kumar | Neeraj Kumar, Swaraj Nanda, Siddharth Singi, Jamal Benhamida, David Kim, Jie-Fu Chen, Amir Momeni-Boroujeni, Gregory M. Goldgof, Gabriele Campanella, and Chad Vanderbilt | Single GPU Task Adaptation of Pathology Foundation Models for Whole Slide Image Analysis | null | null | null | null | cs.CV | http://creativecommons.org/licenses/by/4.0/ | Pathology foundation models (PFMs) have emerged as powerful tools for analyzing whole slide images (WSIs). However, adapting these pretrained PFMs for specific clinical tasks presents considerable challenges, primarily due to the availability of only weak (WSI-level) labels for gigapixel images, necessitating multiple instance learning (MIL) paradigm for effective WSI analysis. This paper proposes a novel approach for single-GPU \textbf{T}ask \textbf{A}daptation of \textbf{PFM}s (TAPFM) that uses vision transformer (\vit) attention for MIL aggregation while optimizing both for feature representations and attention weights. The proposed approach maintains separate computational graphs for MIL aggregator and the PFM to create stable training dynamics that align with downstream task objectives during end-to-end adaptation. Evaluated on mutation prediction tasks for bladder cancer and lung adenocarcinoma across institutional and TCGA cohorts, TAPFM consistently outperforms conventional approaches, with H-Optimus-0 (TAPFM) outperforming the benchmarks. TAPFM effectively handles multi-label classification of actionable mutations as well. Thus, TAPFM makes adaptation of powerful pre-trained PFMs practical on standard hardware for various clinical applications. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 15:56:45 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Kumar",
"Neeraj",
""
],
[
"Nanda",
"Swaraj",
""
],
[
"Singi",
"Siddharth",
""
],
[
"Benhamida",
"Jamal",
""
],
[
"Kim",
"David",
""
],
[
"Chen",
"Jie-Fu",
""
],
[
"Momeni-Boroujeni",
"Amir",
""
],
[
"Goldgof",
"Gregory M.",
""
],
[
"Campanella",
"Gabriele",
""
],
[
"Vanderbilt",
"Chad",
""
]
] |
2506.05185 | Florian Grundbacher | Ferenc Fodor and Florian Grundbacher | On the minimal area of quadrangles circumscribed about planar convex bodies | 11 pages, 3 figures | null | null | null | math.MG | http://creativecommons.org/licenses/by/4.0/ | We show that every planar convex body is contained in a quadrangle whose area is less than $(1 - 2.6 \cdot 10^{-7}) \sqrt{2}$ times the area of the original convex body, improving the best known upper bound by W. Kuperberg. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 15:57:49 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Fodor",
"Ferenc",
""
],
[
"Grundbacher",
"Florian",
""
]
] |
2506.05186 | Ilane Schroetter | I. Schroetter, O. Bern\'e, J. R. Goicoechea, J. H. Black, O. Roncero, F. Alarcon, P. Amiot, O. Asvany, C. Boersma, S. Br\"unken, J. Cami, L. Coudert, E. Dartois, A. Fuente, B. Gans, A. Gusdorf, U. Jacovella, M. A. Martin Drumel, T. Onaka, E. Peeters, E. Roueff, A. G. G. M. Tielens, M. Zannese | PDRs4All XIV: CH radical and $H_3^+$ molecular ion in the irradiated protoplanetary disk d203-506 | 10 pages, 7 figures, submitted to A&A Letters | null | null | null | astro-ph.GA | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Most protoplanetary disks experience a phase in which they are subjected to strong ultraviolet radiation from nearby massive stars. This UV radiation can substantially alter their chemistry by producing numerous radicals and molecular ions. In this Letter we present detailed analysis of the JWST-NIRSpec spectrum of the d203-506 obtained as part of the PDRs4All Early Release Science program. Using state-of-the-art spectroscopic data, we searched for species using a multi-molecule fitting tool, PAHTATmol, that we developed for this purpose. Based on this analysis, we report the clear detection of ro-vibrational emission of the CH radical and likely detection of the H$_3^+$ molecular ion, with estimated abundances of a few times 10$^{-7}$ and approximately 10$^{-8}$, respectively. The presence of CH is predicted by gas-phase models and well explained by hydrocarbon photochemistry. H$_3^+$ is usually formed through reactions of H$_2$ with H$_2^+$ originating from cosmic ray ionization of H$_2$. However, recent theoretical studies suggest that H$_3^+$ also forms through UV-driven chemistry in strongly irradiated ($G_0>$10$^3$), dense ($n_{\rm H} >10^{6}$ cm$^{-3}$) gas. The latter is favored as an explanation for the presence of ``hot'' H$_3^+$ ($T_{\rm ex}\gtrsim$1000 K) in the outer disk layers of d203-506, coinciding with the emission of FUV-pumped H$_2$ and other ``PDR species'', such as CH$^+$, CH$_3^+$, and OH. Our detection of infrared emission from vibrationally excited H$_3^+$ and CH raises questions about their excitation mechanisms and, underscore that UV radiation can have a profound impact on the chemistry of planet forming disks. They also demonstrate the power of JWST pushing the limit for the detection of elusive species in protoplanetary disks. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 15:59:33 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Schroetter",
"I.",
""
],
[
"Berné",
"O.",
""
],
[
"Goicoechea",
"J. R.",
""
],
[
"Black",
"J. H.",
""
],
[
"Roncero",
"O.",
""
],
[
"Alarcon",
"F.",
""
],
[
"Amiot",
"P.",
""
],
[
"Asvany",
"O.",
""
],
[
"Boersma",
"C.",
""
],
[
"Brünken",
"S.",
""
],
[
"Cami",
"J.",
""
],
[
"Coudert",
"L.",
""
],
[
"Dartois",
"E.",
""
],
[
"Fuente",
"A.",
""
],
[
"Gans",
"B.",
""
],
[
"Gusdorf",
"A.",
""
],
[
"Jacovella",
"U.",
""
],
[
"Drumel",
"M. A. Martin",
""
],
[
"Onaka",
"T.",
""
],
[
"Peeters",
"E.",
""
],
[
"Roueff",
"E.",
""
],
[
"Tielens",
"A. G. G. M.",
""
],
[
"Zannese",
"M.",
""
]
] |
2506.05187 | Pierre Pocreau | Alastair A. Abbott, Mehdi Mhalla, Pierre Pocreau | Classical and Quantum Query Complexity of Boolean Functions under Indefinite Causal Order | 13+18 pages, 1+1 figures | null | null | null | quant-ph | http://creativecommons.org/licenses/by/4.0/ | Computational models typically assume that operations are applied in a fixed sequential order. In recent years several works have looked at relaxing this assumption, considering computations without any fixed causal structure and showing that such ''causally indefinite'' computations can provide advantages in various tasks. Recently, the quantum query complexity of Boolean functions has been used as a tool to probe their computational power in a standard complexity theoretic framework, but no separation in exact query complexity has thus-far been found. In this paper, we investigate this problem starting with the simpler and fully classical notion of deterministic query complexity of Boolean functions, and using classical-deterministic processes -- which may exhibit causal indefiniteness -- as a generalised computational framework. We first show that the standard polynomial and certificate lower bounds of deterministic query complexity also hold in such generalised models. Then, we formulate a Boolean function for which causal indefiniteness permits a reduction in query complexity and show that this advantage can be amplified into a polynomial separation. Finally, with the insights gained in the classical-deterministic setting, we give a Boolean function whose quantum query complexity is reduced by causally indefinite computations. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 16:00:36 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Abbott",
"Alastair A.",
""
],
[
"Mhalla",
"Mehdi",
""
],
[
"Pocreau",
"Pierre",
""
]
] |
2506.05188 | Moritz Miller | Moritz Miller and Bernhard Sch\"olkopf and Siyuan Guo | Counterfactual reasoning: an analysis of in-context emergence | null | null | null | null | cs.CL cs.AI cs.LG math.ST stat.TH | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Large-scale neural language models (LMs) exhibit remarkable performance in in-context learning: the ability to learn and reason the input context on the fly without parameter update. This work studies in-context counterfactual reasoning in language models, that is, to predict the consequences of changes under hypothetical scenarios. We focus on studying a well-defined synthetic setup: a linear regression task that requires noise abduction, where accurate prediction is based on inferring and copying the contextual noise from factual observations. We show that language models are capable of counterfactual reasoning in this controlled setup and provide insights that counterfactual reasoning for a broad class of functions can be reduced to a transformation on in-context observations; we find self-attention, model depth, and data diversity in pre-training drive performance in Transformers. More interestingly, our findings extend beyond regression tasks and show that Transformers can perform noise abduction on sequential data, providing preliminary evidence on the potential for counterfactual story generation. Our code is available under https://github.com/moXmiller/counterfactual-reasoning.git . | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 16:02:07 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Miller",
"Moritz",
""
],
[
"Schölkopf",
"Bernhard",
""
],
[
"Guo",
"Siyuan",
""
]
] |
2506.05189 | J. R. Goicoechea | Javier R. Goicoechea, Octavio Roncero, Evelyne Roueff, John H. Black, Ilane Schroetter, Olivier Bern\'e | H3+ in irradiated protoplanetary disks: Linking far-ultraviolet radiation and hot water vapor | Submitted to A&A. Comments are welcome | null | null | null | astro-ph.GA | http://creativecommons.org/licenses/by/4.0/ | The likely JWST detection of vibrationally excited H3+ emission in the irradiated disk d203-506, reported by Schroetter et al., raises the question of whether cosmic-ray ionization is enhanced in disks within clustered star-forming regions, or if alternative mechanisms contribute to H3+ formation and excitation. We present a detailed model of the photodissociation region (PDR) component of a protoplanetary disk-comprising the outer disk surface and the photoevaporative wind-exposed to strong external far-ultraviolet (FUV) radiation. We investigate key gas-phase reactions involving excited H2 that lead to the formation of H3+ in the PDR, including detailed state-to-state dynamical calculations of reactions H2 + HOC+ -> H3+ + CO and H2 + H+ -> H2+ + H. We also consider the effects of photoionization of vibrationally excited H2(v>=4), a process that has not previously been included in disk models. We find that these FUV-driven reactions dominate the formation of H3+ in the PDR of strongly irradiated disks, largely independently of cosmic-ray ionization. The predicted H3+ abundance peaks at x(H3+)~10^-8, coinciding with regions where HOC+ is also abundant. The high abundances of H3+ and HOC+ are ultimately linked to the strength of the external FUV field (G0), the presence of C+, and an enhanced reservoir of hot H2O. The predicted H3+ column densities (~10^13 cm^-2) are consistent with the presence of H3+ in the PDR of irradiated disks. We also find that formation pumping, resulting from exoergic reactions between excited H2 and HOC+, drive the vibrational excitation of H3+. We expect this photochemistry to be highly active in disks where G0 > 10^3. The H3+ formation pathways studied here may also be relevant in the inner disk region (near the host star), in exoplanetary ionospheres, and in the early Universe. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 16:02:11 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Goicoechea",
"Javier R.",
""
],
[
"Roncero",
"Octavio",
""
],
[
"Roueff",
"Evelyne",
""
],
[
"Black",
"John H.",
""
],
[
"Schroetter",
"Ilane",
""
],
[
"Berné",
"Olivier",
""
]
] |
2506.05190 | Chris Kapulkin | Daniel Carranza, Chris Kapulkin, Nathan Kershaw, Reinhard Laubenbacher, Matthew Wheeler | Categorical foundations of discrete dynamical systems | 30 pages; comments welcome | null | null | null | math.DS math.CT | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | We develop categorical foundations of discrete dynamical systems, aimed at understanding how the structure of the system affects its dynamics. The key technical innovation is the notion of a cycle set, which provides a formal language in which to speak of the system's attractors. As a proof of concept, we provide a decomposition theorem for discrete dynamical systems. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 16:03:40 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Carranza",
"Daniel",
""
],
[
"Kapulkin",
"Chris",
""
],
[
"Kershaw",
"Nathan",
""
],
[
"Laubenbacher",
"Reinhard",
""
],
[
"Wheeler",
"Matthew",
""
]
] |
2506.05191 | Yake Wei | Yake Wei, Yu Miao, Dongzhan Zhou, Di Hu | MokA: Multimodal Low-Rank Adaptation for MLLMs | null | null | null | null | cs.CV | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | In this paper, we reveal that most current efficient multimodal fine-tuning methods are hindered by a key limitation: they are directly borrowed from LLMs, often neglecting the intrinsic differences of multimodal scenarios and even affecting the full utilization of all modalities. Inspired by our empirical observation, we argue that unimodal adaptation and cross-modal adaptation are two essential parts for the effective fine-tuning of MLLMs. From this perspective, we propose Multimodal low-rank Adaptation (MokA), a multimodal-aware efficient fine-tuning strategy that takes multimodal characteristics into consideration. It compresses unimodal information by modality-specific parameters while explicitly enhancing cross-modal interaction, ensuring both unimodal and cross-modal adaptation. Extensive experiments cover three representative multimodal scenarios (audio-visual-text, visual-text, and speech-text), and multiple LLM backbones (LLaMA2/3, Qwen2, Qwen2.5-VL, etc). Consistent improvements indicate the efficacy and versatility of the proposed method. Ablation studies and efficiency evaluation are also conducted to fully asses our method. Overall, we think MokA provides a more targeted solution for efficient adaptation of MLLMs, paving the way for further exploration. The project page is at https://gewu-lab.github.io/MokA. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 16:04:08 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Wei",
"Yake",
""
],
[
"Miao",
"Yu",
""
],
[
"Zhou",
"Dongzhan",
""
],
[
"Hu",
"Di",
""
]
] |
2506.05192 | Johannes Lehmann | Christel Baier, Rio Klatt, Sascha Kl\"uppelholz, Johannes Lehmann | Backward Responsibility in Transition Systems Beyond Safety | null | null | null | null | cs.FL | http://creativecommons.org/licenses/by-sa/4.0/ | As the complexity of software systems rises, methods for explaining their behaviour are becoming ever-more important. When a system fails, it is critical to determine which of its components are responsible for this failure. Within the verification community, one approach uses graph games and the Shapley value to ascribe a responsibility value to every state of a transition system. As this is done with respect to a specific failure, it is called backward responsibility.
This paper provides tight complexity bounds for backward responsibility for reachability, B\"uchi and parity objectives. For B\"uchi objectives, a polynomial algorithm is given to determine the set of responsible states. To analyse systems that are too large for standard methods, the paper presents a novel refinement algorithm that iteratively computes responsibility and demonstrates its utility with a prototypical implementation. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 16:04:41 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Baier",
"Christel",
""
],
[
"Klatt",
"Rio",
""
],
[
"Klüppelholz",
"Sascha",
""
],
[
"Lehmann",
"Johannes",
""
]
] |
2506.05193 | Hongmiao Yu | Hongmiao Yu | On the weak and strong Lefschetz properties for initial ideals of determinantal ideals with respect to diagonal monomial orders | 30 pages, 12 figures | null | null | null | math.AC | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | We study the weak and strong Lefschetz properties for $R/\mathrm{in}(I_t)$, where $I_t$ is the ideal of a polynomial ring $R$ generated by the $t$-minors of an $m\times n$ matrix of indeterminates, and $\mathrm{in}(I_t)$ denotes the initial ideal of $I_t$ with respect to a diagonal monomial order. We show that when $I_t$ is generated by maximal minors (that is, $t=\mathrm{min}\{m,n\}$), the ring $R/\mathrm{in}(I_t)$ has the strong Lefschetz property for all $m$, $n$. In contrast, for $t<\mathrm{min}\{m,n\}$, we provide a bound such that $R/\mathrm{in}(I_t)$ fails to satisfy the weak Lefschetz property whenever the product $mn$ exceeds this bound. As an application, we present counterexamples that provide a negative answer to a question posed by Murai regarding the preservation of Lefschetz properties under square-free Gr\"obner degenerations. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 16:05:45 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Yu",
"Hongmiao",
""
]
] |
2506.05194 | Viktor Harangi | Viktor Harangi | Star decompositions via orientations | null | null | null | null | math.CO math.PR | http://creativecommons.org/licenses/by/4.0/ | A $k$-star decomposition of a graph is a partition of its edges into $k$-stars (i.e., $k$ edges with a common vertex). The paper studies the following problem: given $k \leq d/2$, does the random $d$-regular graph have a $k$-star decomposition (asymptotically almost surely, provided that the number of edges is divisible by $k$)? Delcourt, Greenhill, Isaev, Lidick\'y, and Postle proved the a.a.s. existence for every odd $k$ using earlier results regarding orientations satisfying certain degree conditions modulo $k$.
In this paper we give a direct, self-contained proof that works for every $d$ and every $k<d/2-1$. In fact, we prove stronger results. Let $s\geq 1$ denote the integer part of $d/(2k)$. We show that the random $d$-regular graph a.a.s. has a $k$-star decomposition such that the number of stars centered at each vertex is either $s$ or $s+1$. Moreover, if $k < d/3$ or $k \leq d/2 - 2.6 \log d$, we can even prescribe the set of vertices with $s$ stars, as long as it is of the appropriate size. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 16:07:20 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Harangi",
"Viktor",
""
]
] |
2506.05195 | Mohammed Eltayeb | Josue Marroquin, Nan Inzali, Miles Dillon Lantz, Campbell Freeman, Amod Ashtekar, \\Ajinkya Umesh Mulik, and Mohammed E Eltayeb | Vision-Based Autonomous MM-Wave Reflector Using ArUco-Driven Angle-of-Arrival Estimation | null | null | null | null | cs.CV | http://creativecommons.org/licenses/by/4.0/ | Reliable millimeter-wave (mmWave) communication in non-line-of-sight (NLoS) conditions remains a major challenge for both military and civilian operations, especially in urban or infrastructure-limited environments. This paper presents a vision-aided autonomous reflector system designed to enhance mmWave link performance by dynamically steering signal reflections using a motorized metallic plate. The proposed system leverages a monocular camera to detect ArUco markers on allied transmitter and receiver nodes, estimate their angles of arrival, and align the reflector in real time for optimal signal redirection. This approach enables selective beam coverage by serving only authenticated targets with visible markers and reduces the risk of unintended signal exposure. The designed prototype, built on a Raspberry Pi 4 and low-power hardware, operates autonomously without reliance on external infrastructure or GPS. Experimental results at 60\,GHz demonstrate a 23\,dB average gain in received signal strength and an 0.89 probability of maintaining signal reception above a target threshold of -65 dB in an indoor environment, far exceeding the static and no-reflector baselines. These results demonstrate the system's potential for resilient and adaptive mmWave connectivity in complex and dynamic environments. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 16:07:22 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Marroquin",
"Josue",
""
],
[
"Inzali",
"Nan",
""
],
[
"Lantz",
"Miles Dillon",
""
],
[
"Freeman",
"Campbell",
""
],
[
"Ashtekar",
"Amod",
""
],
[
"Mulik",
"\\\\Ajinkya Umesh",
""
],
[
"Eltayeb",
"Mohammed E",
""
]
] |
2506.05196 | Jifei Luo | Jifei Luo, Wenzheng Wu, Hantao Yao, Lu Yu, Changsheng Xu | Locality Preserving Markovian Transition for Instance Retrieval | This paper has been accepted by ICML2025 | null | null | null | cs.LG | http://creativecommons.org/licenses/by/4.0/ | Diffusion-based re-ranking methods are effective in modeling the data manifolds through similarity propagation in affinity graphs. However, positive signals tend to diminish over several steps away from the source, reducing discriminative power beyond local regions. To address this issue, we introduce the Locality Preserving Markovian Transition (LPMT) framework, which employs a long-term thermodynamic transition process with multiple states for accurate manifold distance measurement. The proposed LPMT first integrates diffusion processes across separate graphs using Bidirectional Collaborative Diffusion (BCD) to establish strong similarity relationships. Afterwards, Locality State Embedding (LSE) encodes each instance into a distribution for enhanced local consistency. These distributions are interconnected via the Thermodynamic Markovian Transition (TMT) process, enabling efficient global retrieval while maintaining local effectiveness. Experimental results across diverse tasks confirm the effectiveness of LPMT for instance retrieval. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 16:07:31 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Luo",
"Jifei",
""
],
[
"Wu",
"Wenzheng",
""
],
[
"Yao",
"Hantao",
""
],
[
"Yu",
"Lu",
""
],
[
"Xu",
"Changsheng",
""
]
] |
2506.05197 | Gianmarco Perrupato | Marco Benedetti, Andrej Bogdanov, Enrico M. Malatesta, Marc M\'ezard, Gianmarco Perrupato, Alon Rosen, Nikolaj I. Schwartzbach, Riccardo Zecchina | Overlap Gap and Computational Thresholds in the Square Wave Perceptron | 28 pages, 20 figures | null | null | null | cond-mat.dis-nn | http://creativecommons.org/licenses/by/4.0/ | Square Wave Perceptrons (SWPs) form a class of neural network models with oscillating activation function that exhibit intriguing ``hardness'' properties in the high-dimensional limit at a fixed signal-to-noise ratio $\alpha = O(1)$. In this work, we examine two key aspects of these models. The first is related to the so-called overlap-gap property, that is a disconnectivity feature of the geometry of the solution space of combinatorial optimization problems proven to cause the failure of a large family of solvers, and conjectured to be a symptom of algorithmic hardness. We identify, both in the storage and in the teacher-student settings, the emergence of an overlap gap at a threshold $\alpha_{\mathrm{OGP}}(\delta)$, which can be made arbitrarily small by suitably increasing the frequency of oscillations $1/\delta$ of the activation. This suggests that in this small-$\delta$ regime, typical instances of the problem are hard to solve even for small values of $\alpha$. Second, in the teacher-student setup, we show that the recovery threshold of the planted signal for message-passing algorithms can be made arbitrarily large by reducing $\delta$. These properties make SWPs both a challenging benchmark for algorithms and an interesting candidate for cryptographic applications. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 16:10:23 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Benedetti",
"Marco",
""
],
[
"Bogdanov",
"Andrej",
""
],
[
"Malatesta",
"Enrico M.",
""
],
[
"Mézard",
"Marc",
""
],
[
"Perrupato",
"Gianmarco",
""
],
[
"Rosen",
"Alon",
""
],
[
"Schwartzbach",
"Nikolaj I.",
""
],
[
"Zecchina",
"Riccardo",
""
]
] |
2506.05198 | Yuxin Wen | Yuxin Wen, Yangsibo Huang, Tom Goldstein, Ravi Kumar, Badih Ghazi, Chiyuan Zhang | Quantifying Cross-Modality Memorization in Vision-Language Models | null | null | null | null | cs.CV cs.LG | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Understanding what and how neural networks memorize during training is crucial, both from the perspective of unintentional memorization of potentially sensitive information and from the standpoint of effective knowledge acquisition for real-world, knowledge-intensive tasks. While previous studies primarily investigate memorization within a single modality, such as text memorization in large language models or image memorization in diffusion models, unified multimodal models are becoming increasingly prevalent in practical applications. In this work, we focus on the unique characteristics of cross-modality memorization and conduct a systematic study centered on vision-language models. To facilitate controlled experiments, we first introduce a synthetic persona dataset comprising diverse synthetic person images and textual descriptions. We quantify factual knowledge memorization and cross-modal transferability by training models on a single modality and evaluating their performance in the other. Our results reveal that facts learned in one modality transfer to the other, but a significant gap exists between recalling information in the source and target modalities. Furthermore, we observe that this gap exists across various scenarios, including more capable models, machine unlearning, and the multi-hop case. At the end, we propose a baseline method to mitigate this challenge. We hope our study can inspire future research on developing more robust multimodal learning techniques to enhance cross-modal transferability. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 16:10:47 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Wen",
"Yuxin",
""
],
[
"Huang",
"Yangsibo",
""
],
[
"Goldstein",
"Tom",
""
],
[
"Kumar",
"Ravi",
""
],
[
"Ghazi",
"Badih",
""
],
[
"Zhang",
"Chiyuan",
""
]
] |
2506.05199 | Yani Zhang | Yani Zhang, Dongming Wu, Hao Shi, Yingfei Liu, Tiancai Wang, Haoqiang Fan, Xingping Dong | Grounding Beyond Detection: Enhancing Contextual Understanding in Embodied 3D Grounding | 1st place on embodiedscan | null | null | null | cs.CV | http://creativecommons.org/licenses/by/4.0/ | Embodied 3D grounding aims to localize target objects described in human instructions from ego-centric viewpoint. Most methods typically follow a two-stage paradigm where a trained 3D detector's optimized backbone parameters are used to initialize a grounding model. In this study, we explore a fundamental question: Does embodied 3D grounding benefit enough from detection? To answer this question, we assess the grounding performance of detection models using predicted boxes filtered by the target category. Surprisingly, these detection models without any instruction-specific training outperform the grounding models explicitly trained with language instructions. This indicates that even category-level embodied 3D grounding may not be well resolved, let alone more fine-grained context-aware grounding. Motivated by this finding, we propose DEGround, which shares DETR queries as object representation for both DEtection and Grounding and enables the grounding to benefit from basic category classification and box detection. Based on this framework, we further introduce a regional activation grounding module that highlights instruction-related regions and a query-wise modulation module that incorporates sentence-level semantic into the query representation, strengthening the context-aware understanding of language instructions. Remarkably, DEGround outperforms state-of-the-art model BIP3D by 7.52\% at overall accuracy on the EmbodiedScan validation set. The source code will be publicly available at https://github.com/zyn213/DEGround. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 16:11:57 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Zhang",
"Yani",
""
],
[
"Wu",
"Dongming",
""
],
[
"Shi",
"Hao",
""
],
[
"Liu",
"Yingfei",
""
],
[
"Wang",
"Tiancai",
""
],
[
"Fan",
"Haoqiang",
""
],
[
"Dong",
"Xingping",
""
]
] |
2506.05200 | Yuchen Jiao | Gen Li, Yuchen Jiao, Yu Huang, Yuting Wei, Yuxin Chen | Transformers Meet In-Context Learning: A Universal Approximation Theory | null | null | null | null | cs.LG math.ST stat.ML stat.TH | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Modern large language models are capable of in-context learning, the ability to perform new tasks at inference time using only a handful of input-output examples in the prompt, without any fine-tuning or parameter updates. We develop a universal approximation theory to better understand how transformers enable in-context learning. For any class of functions (each representing a distinct task), we demonstrate how to construct a transformer that, without any further weight updates, can perform reliable prediction given only a few in-context examples. In contrast to much of the recent literature that frames transformers as algorithm approximators -- i.e., constructing transformers to emulate the iterations of optimization algorithms as a means to approximate solutions of learning problems -- our work adopts a fundamentally different approach rooted in universal function approximation. This alternative approach offers approximation guarantees that are not constrained by the effectiveness of the optimization algorithms being approximated, thereby extending far beyond convex problems and linear function classes. Our construction sheds light on how transformers can simultaneously learn general-purpose representations and adapt dynamically to in-context examples. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 16:12:51 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Li",
"Gen",
""
],
[
"Jiao",
"Yuchen",
""
],
[
"Huang",
"Yu",
""
],
[
"Wei",
"Yuting",
""
],
[
"Chen",
"Yuxin",
""
]
] |
2506.05201 | Felix Wilsch | Sabine Kraml, Andre Lessa, Suraj Prakash, Felix Wilsch | SUSY meets SMEFT: Complete one-loop matching of the general MSSM | 54 pages, 3 tables, 3 figures, auxiliary material available on GitHub: https://github.com/BSM-EFT/MSSM-to-SMEFT | null | null | TTK-25-14, P3H-25-036 | hep-ph | http://creativecommons.org/licenses/by/4.0/ | We present the complete one-loop matching of the Minimal Supersymmetric Standard Model (MSSM) onto the Standard Model Effective Field Theory (SMEFT), considering the most general case for the MSSM with conserved $R$-parity, which has 124 free parameters. The matching is performed with the Matchete package, which integrates out all superpartners at once with non-degenerate masses, while also retaining the most general flavor structure. Our results include all correlations among the different SMEFT Wilson coefficients that are governed by supersymmetry and thus provide a basis for future systematic and global studies of the MSSM parameter space employing EFT methods. A detailed discussion is provided on the treatment of the Higgs sector and electroweak symmetry breaking, along with the reduction of redundant operators in the EFT Lagrangian to the Warsaw basis. Furthermore, we validate against existing results in the literature and present a minimal phenomenological example. Extensive auxiliary material, including the code utilized for the matching, is available on GitHub. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 16:14:25 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Kraml",
"Sabine",
""
],
[
"Lessa",
"Andre",
""
],
[
"Prakash",
"Suraj",
""
],
[
"Wilsch",
"Felix",
""
]
] |
2506.05202 | Daniele Tramontano | Daniele Tramontano, Yaroslav Kivva, Saber Salehkaleybar Mathias Drton, Negar Kiyavash | Causal Effect Identification in lvLiNGAM from Higher-Order Cumulants | Accepted at ICML 2025 | null | null | null | stat.ML cs.LG stat.ME | http://creativecommons.org/licenses/by/4.0/ | This paper investigates causal effect identification in latent variable Linear Non-Gaussian Acyclic Models (lvLiNGAM) using higher-order cumulants, addressing two prominent setups that are challenging in the presence of latent confounding: (1) a single proxy variable that may causally influence the treatment and (2) underspecified instrumental variable cases where fewer instruments exist than treatments. We prove that causal effects are identifiable with a single proxy or instrument and provide corresponding estimation methods. Experimental results demonstrate the accuracy and robustness of our approaches compared to existing methods, advancing the theoretical and practical understanding of causal inference in linear systems with latent confounders. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 16:14:35 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Tramontano",
"Daniele",
""
],
[
"Kivva",
"Yaroslav",
""
],
[
"Drton",
"Saber Salehkaleybar Mathias",
""
],
[
"Kiyavash",
"Negar",
""
]
] |
2506.05203 | Leonardo Ceragioli | Leonardo Ceragioli and Giuseppe Primiero | Trustworthiness Preservation by Copies of Machine Learning Systems | null | null | null | null | cs.LO cs.LG | http://creativecommons.org/licenses/by/4.0/ | A common practice of ML systems development concerns the training of the same model under different data sets, and the use of the same (training and test) sets for different learning models. The first case is a desirable practice for identifying high quality and unbiased training conditions. The latter case coincides with the search for optimal models under a common dataset for training. These differently obtained systems have been considered akin to copies. In the quest for responsible AI, a legitimate but hardly investigated question is how to verify that trustworthiness is preserved by copies. In this paper we introduce a calculus to model and verify probabilistic complex queries over data and define four distinct notions: Justifiably, Equally, Weakly and Almost Trustworthy which can be checked analysing the (partial) behaviour of the copy with respect to its original. We provide a study of the relations between these notions of trustworthiness, and how they compose with each other and under logical operations. The aim is to offer a computational tool to check the trustworthiness of possibly complex systems copied from an original whose behavour is known. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 16:14:57 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Ceragioli",
"Leonardo",
""
],
[
"Primiero",
"Giuseppe",
""
]
] |
2506.05204 | Yanbo Wang | Yanbo Wang, Ziyi Wang, Wenzhao Zheng, Jie Zhou and Jiwen Lu | OGGSplat: Open Gaussian Growing for Generalizable Reconstruction with Expanded Field-of-View | null | null | null | null | cs.CV | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Reconstructing semantic-aware 3D scenes from sparse views is a challenging yet essential research direction, driven by the demands of emerging applications such as virtual reality and embodied AI. Existing per-scene optimization methods require dense input views and incur high computational costs, while generalizable approaches often struggle to reconstruct regions outside the input view cone. In this paper, we propose OGGSplat, an open Gaussian growing method that expands the field-of-view in generalizable 3D reconstruction. Our key insight is that the semantic attributes of open Gaussians provide strong priors for image extrapolation, enabling both semantic consistency and visual plausibility. Specifically, once open Gaussians are initialized from sparse views, we introduce an RGB-semantic consistent inpainting module applied to selected rendered views. This module enforces bidirectional control between an image diffusion model and a semantic diffusion model. The inpainted regions are then lifted back into 3D space for efficient and progressive Gaussian parameter optimization. To evaluate our method, we establish a Gaussian Outpainting (GO) benchmark that assesses both semantic and generative quality of reconstructed open-vocabulary scenes. OGGSplat also demonstrates promising semantic-aware scene reconstruction capabilities when provided with two view images captured directly from a smartphone camera. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 16:17:18 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Wang",
"Yanbo",
""
],
[
"Wang",
"Ziyi",
""
],
[
"Zheng",
"Wenzhao",
""
],
[
"Zhou",
"Jie",
""
],
[
"Lu",
"Jiwen",
""
]
] |
2506.05205 | Jackson Petty | Jackson Petty, Michael Y. Hu, Wentao Wang, Shauli Ravfogel, William Merrill, Tal Linzen | RELIC: Evaluating Compositional Instruction Following via Language Recognition | null | null | null | null | cs.CL | http://creativecommons.org/licenses/by/4.0/ | Large language models (LLMs) are increasingly expected to perform tasks based only on a specification of the task provided in context, without examples of inputs and outputs; this ability is referred to as instruction following. We introduce the Recognition of Languages In-Context (RELIC) framework to evaluate instruction following using language recognition: the task of determining if a string is generated by formal grammar. Unlike many standard evaluations of LLMs' ability to use their context, this task requires composing together a large number of instructions (grammar productions) retrieved from the context. Because the languages are synthetic, the task can be increased in complexity as LLMs' skills improve, and new instances can be automatically generated, mitigating data contamination. We evaluate state-of-the-art LLMs on RELIC and find that their accuracy can be reliably predicted from the complexity of the grammar and the individual example strings, and that even the most advanced LLMs currently available show near-chance performance on more complex grammars and samples, in line with theoretical expectations. We also use RELIC to diagnose how LLMs attempt to solve increasingly difficult reasoning tasks, finding that as the complexity of the language recognition task increases, models switch to relying on shallow heuristics instead of following complex instructions. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 16:17:24 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Petty",
"Jackson",
""
],
[
"Hu",
"Michael Y.",
""
],
[
"Wang",
"Wentao",
""
],
[
"Ravfogel",
"Shauli",
""
],
[
"Merrill",
"William",
""
],
[
"Linzen",
"Tal",
""
]
] |
2506.05206 | Bar Alon | Bar Alon, Moshe Goldstein, Roni Ilan | The non-Hermitian magnetic moment | null | null | null | null | cond-mat.mes-hall physics.optics quant-ph | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | We construct a semiclassical theory for electrons in a non-Hermitian periodic system subject to perturbations varying slowly in space and time. We derive the energy of the wavepacket to first order in the gradients of the perturbations. Applying the theory to the specific case of a uniform external magnetic field, we obtain an expression for the orbital magnetization energy. Using the principles of non-Hermitian dynamics, we define a physically meaningful non-Hermitian generalization of the angular momentum operator and show that it is compatible with the real part of the orbital magnetic moment. The imaginary part of the orbital magnetic moment is also discussed and shown to originate from an imaginary counterpart to the angular momentum that gives rise to a non-Hermitian generalization of the Aharonov-Bohm effect. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 16:18:10 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Alon",
"Bar",
""
],
[
"Goldstein",
"Moshe",
""
],
[
"Ilan",
"Roni",
""
]
] |
2506.05207 | Kunyu Feng | Yue Ma, Yulong Liu, Qiyuan Zhu, Ayden Yang, Kunyu Feng, Xinhua Zhang, Zhifeng Li, Sirui Han, Chenyang Qi, Qifeng Chen | Follow-Your-Motion: Video Motion Transfer via Efficient Spatial-Temporal Decoupled Finetuning | project page: https://follow-your-motion.github.io/ | null | null | null | cs.CV | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Recently, breakthroughs in the video diffusion transformer have shown remarkable capabilities in diverse motion generations. As for the motion-transfer task, current methods mainly use two-stage Low-Rank Adaptations (LoRAs) finetuning to obtain better performance. However, existing adaptation-based motion transfer still suffers from motion inconsistency and tuning inefficiency when applied to large video diffusion transformers. Naive two-stage LoRA tuning struggles to maintain motion consistency between generated and input videos due to the inherent spatial-temporal coupling in the 3D attention operator. Additionally, they require time-consuming fine-tuning processes in both stages. To tackle these issues, we propose Follow-Your-Motion, an efficient two-stage video motion transfer framework that finetunes a powerful video diffusion transformer to synthesize complex motion.Specifically, we propose a spatial-temporal decoupled LoRA to decouple the attention architecture for spatial appearance and temporal motion processing. During the second training stage, we design the sparse motion sampling and adaptive RoPE to accelerate the tuning speed. To address the lack of a benchmark for this field, we introduce MotionBench, a comprehensive benchmark comprising diverse motion, including creative camera motion, single object motion, multiple object motion, and complex human motion. We show extensive evaluations on MotionBench to verify the superiority of Follow-Your-Motion. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 16:18:32 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Ma",
"Yue",
""
],
[
"Liu",
"Yulong",
""
],
[
"Zhu",
"Qiyuan",
""
],
[
"Yang",
"Ayden",
""
],
[
"Feng",
"Kunyu",
""
],
[
"Zhang",
"Xinhua",
""
],
[
"Li",
"Zhifeng",
""
],
[
"Han",
"Sirui",
""
],
[
"Qi",
"Chenyang",
""
],
[
"Chen",
"Qifeng",
""
]
] |
2506.05208 | Yuhua Zhu | Yuhua Zhu, Yuming Zhang, Haoyu Zhang | Optimal-PhiBE: A PDE-based Model-free framework for Continuous-time Reinforcement Learning | null | null | null | null | math.OC | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | This paper addresses continuous-time reinforcement learning (CTRL) where the system dynamics are governed by a stochastic differential equation but are unknown, and only discrete-time observations are available. Existing approaches face limitations: model-based PDE methods suffer from non-identifiability, while model-free methods based on the optimal Bellman equation (Optimal-BE) are prone to large discretization errors sensitive to both the dynamics and reward structure. To overcome these challenges, we introduce Optimal-PhiBE, a formulation that integrates discrete-time information into a continuous-time PDE, combining the strength of both existing frameworks while mitigating their limitations. Optimal-PhiBE avoids explicit dynamics estimation, exhibits smaller discretization errors when the uncontrolled system evolves slowly, and demonstrates reduced sensitivity to oscillatory reward structures. In the linear-quadratic regulator (LQR) setting, sharp error bounds are established for both Optimal-PhiBE and Optimal-BE. The results show that Optimal-PhiBE exactly recovers the optimal policy in the undiscounted case and substantially outperforms Optimal-BE when the problem is weakly discounted or control-dominant. Furthermore, we extend Optimal-PhiBE to higher orders, providing increasingly accurate approximations. A model-free policy iteration algorithm is proposed to solve the Optimal-PhiBE directly from trajectory data. Numerical experiments are conducted to verify the theoretical findings. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 16:20:15 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Zhu",
"Yuhua",
""
],
[
"Zhang",
"Yuming",
""
],
[
"Zhang",
"Haoyu",
""
]
] |
2506.05209 | Colin Raffel | Nikhil Kandpal, Brian Lester, Colin Raffel, Sebastian Majstorovic, Stella Biderman, Baber Abbasi, Luca Soldaini, Enrico Shippole, A. Feder Cooper, Aviya Skowron, John Kirchenbauer, Shayne Longpre, Lintang Sutawika, Alon Albalak, Zhenlin Xu, Guilherme Penedo, Loubna Ben Allal, Elie Bakouch, John David Pressman, Honglu Fan, Dashiell Stander, Guangyu Song, Aaron Gokaslan, Tom Goldstein, Brian R. Bartoldson, Bhavya Kailkhura, Tyler Murray | The Common Pile v0.1: An 8TB Dataset of Public Domain and Openly Licensed Text | null | null | null | null | cs.CL cs.LG | http://creativecommons.org/licenses/by/4.0/ | Large language models (LLMs) are typically trained on enormous quantities of unlicensed text, a practice that has led to scrutiny due to possible intellectual property infringement and ethical concerns. Training LLMs on openly licensed text presents a first step towards addressing these issues, but prior data collection efforts have yielded datasets too small or low-quality to produce performant LLMs. To address this gap, we collect, curate, and release the Common Pile v0.1, an eight terabyte collection of openly licensed text designed for LLM pretraining. The Common Pile comprises content from 30 sources that span diverse domains including research papers, code, books, encyclopedias, educational materials, audio transcripts, and more. Crucially, we validate our efforts by training two 7 billion parameter LLMs on text from the Common Pile: Comma v0.1-1T and Comma v0.1-2T, trained on 1 and 2 trillion tokens respectively. Both models attain competitive performance to LLMs trained on unlicensed text with similar computational budgets, such as Llama 1 and 2 7B. In addition to releasing the Common Pile v0.1 itself, we also release the code used in its creation as well as the training mixture and checkpoints for the Comma v0.1 models. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 16:21:30 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Kandpal",
"Nikhil",
""
],
[
"Lester",
"Brian",
""
],
[
"Raffel",
"Colin",
""
],
[
"Majstorovic",
"Sebastian",
""
],
[
"Biderman",
"Stella",
""
],
[
"Abbasi",
"Baber",
""
],
[
"Soldaini",
"Luca",
""
],
[
"Shippole",
"Enrico",
""
],
[
"Cooper",
"A. Feder",
""
],
[
"Skowron",
"Aviya",
""
],
[
"Kirchenbauer",
"John",
""
],
[
"Longpre",
"Shayne",
""
],
[
"Sutawika",
"Lintang",
""
],
[
"Albalak",
"Alon",
""
],
[
"Xu",
"Zhenlin",
""
],
[
"Penedo",
"Guilherme",
""
],
[
"Allal",
"Loubna Ben",
""
],
[
"Bakouch",
"Elie",
""
],
[
"Pressman",
"John David",
""
],
[
"Fan",
"Honglu",
""
],
[
"Stander",
"Dashiell",
""
],
[
"Song",
"Guangyu",
""
],
[
"Gokaslan",
"Aaron",
""
],
[
"Goldstein",
"Tom",
""
],
[
"Bartoldson",
"Brian R.",
""
],
[
"Kailkhura",
"Bhavya",
""
],
[
"Murray",
"Tyler",
""
]
] |
2506.05210 | Jan Ackermann | Jan Ackermann, Kiyohiro Nakayama, Guandao Yang, Tong Wu, Gordon Wetzstein | Towards Vision-Language-Garment Models For Web Knowledge Garment Understanding and Generation | Presented at MMFM CVPRW'25, code available at https://georgenakayama.github.io/AIpparel/ | null | null | null | cs.CV | http://creativecommons.org/licenses/by/4.0/ | Multimodal foundation models have demonstrated strong generalization, yet their ability to transfer knowledge to specialized domains such as garment generation remains underexplored. We introduce VLG, a vision-language-garment model that synthesizes garments from textual descriptions and visual imagery. Our experiments assess VLG's zero-shot generalization, investigating its ability to transfer web-scale reasoning to unseen garment styles and prompts. Preliminary results indicate promising transfer capabilities, highlighting the potential for multimodal foundation models to adapt effectively to specialized domains like fashion design. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 16:22:17 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Ackermann",
"Jan",
""
],
[
"Nakayama",
"Kiyohiro",
""
],
[
"Yang",
"Guandao",
""
],
[
"Wu",
"Tong",
""
],
[
"Wetzstein",
"Gordon",
""
]
] |
2506.05211 | David Atkinson | David Atkinson, Jena D. Hwang, and Jacob Morrison | Intentionally Unintentional: GenAI Exceptionalism and the First Amendment | null | null | null | null | cs.CY cs.AI | http://creativecommons.org/licenses/by-nc-sa/4.0/ | This paper challenges the assumption that courts should grant First Amendment protections to outputs from large generative AI models, such as GPT-4 and Gemini. We argue that because these models lack intentionality, their outputs do not constitute speech as understood in the context of established legal precedent, so there can be no speech to protect. Furthermore, if the model outputs are not speech, users cannot claim a First Amendment speech right to receive the outputs. We also argue that extending First Amendment rights to AI models would not serve the fundamental purposes of free speech, such as promoting a marketplace of ideas, facilitating self-governance, or fostering self-expression. In fact, granting First Amendment protections to AI models would be detrimental to society because it would hinder the government's ability to regulate these powerful technologies effectively, potentially leading to the unchecked spread of misinformation and other harms. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 16:26:32 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Atkinson",
"David",
""
],
[
"Hwang",
"Jena D.",
""
],
[
"Morrison",
"Jacob",
""
]
] |
2506.05212 | Philippe Moustrou | Emmanuel Hallouin, Philippe Moustrou and Marc Perret | Refinements on higher order Weil-Oesterl\'e bounds via a Serre type argument | 17 pages, 2 figures | null | null | null | math.NT math.AG | http://creativecommons.org/licenses/by/4.0/ | Weil's theorem gives the most standard bound on the number of points of a curve over a finite field. This bound was improved by Ihara and Oesterl\'e for larger genus. Recently, Hallouin and Perret gave a new point of view on these bounds, that can be obtained by solving a sequence of semi-definite programs, and the two first steps of this hierarchy recover Weil's and Ihara's bounds. On the other hand, by taking into account arithmetic constraints, Serre obtained a refinement on Weil's bound. In this article, we combine these two approaches and propose a strengthening of Ihara's bound, based on an argument similar to Serre's refinement. We show that this generically improves upon Ihara's bound, even in the range where it was the best bound so far. Finally we discuss possible extensions to higher order Weil-Oesterl\'e bounds. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 16:27:10 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Hallouin",
"Emmanuel",
""
],
[
"Moustrou",
"Philippe",
""
],
[
"Perret",
"Marc",
""
]
] |
2506.05213 | Nathan Herr | Nathan Herr, Tim Rockt\"aschel, Roberta Raileanu | LLM-First Search: Self-Guided Exploration of the Solution Space | 9 main pages, 2 figures, 2 tables, 36 appendix pages | null | null | null | cs.AI cs.CL | http://creativecommons.org/licenses/by/4.0/ | Large Language Models (LLMs) have demonstrated remarkable improvements in reasoning and planning through increased test-time compute, often by framing problem-solving as a search process. While methods like Monte Carlo Tree Search (MCTS) have proven effective in some domains, their reliance on fixed exploration hyperparameters limits their adaptability across tasks of varying difficulty, rendering them impractical or expensive in certain settings. In this paper, we propose \textbf{LLM-First Search (LFS)}, a novel \textit{LLM Self-Guided Search} method that removes the need for pre-defined search strategies by empowering the LLM to autonomously control the search process via self-guided exploration. Rather than relying on external heuristics or hardcoded policies, the LLM evaluates whether to pursue the current search path or explore alternative branches based on its internal scoring mechanisms. This enables more flexible and context-sensitive reasoning without requiring manual tuning or task-specific adaptation. We evaluate LFS on Countdown and Sudoku against three classic widely-used search algorithms, Tree-of-Thoughts' Breadth First Search (ToT-BFS), Best First Search (BestFS), and MCTS, each of which have been used to achieve SotA results on a range of challenging reasoning tasks. We found that LFS (1) performs better on more challenging tasks without additional tuning, (2) is more computationally efficient compared to the other methods, especially when powered by a stronger model, (3) scales better with stronger models, due to its LLM-First design, and (4) scales better with increased compute budget. Our code is publicly available at \href{https://github.com/NathanHerr/LLM-First-Search}{LLM-First-Search}. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 16:27:49 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Herr",
"Nathan",
""
],
[
"Rocktäschel",
"Tim",
""
],
[
"Raileanu",
"Roberta",
""
]
] |
2506.05214 | Jingyu Hu | Jingyu Hu, Hongbo Bo, Jun Hong, Xiaowei Liu, Weiru Liu | Mitigating Degree Bias Adaptively with Hard-to-Learn Nodes in Graph Contrastive Learning | null | null | null | null | cs.LG cs.AI cs.CL | http://creativecommons.org/licenses/by/4.0/ | Graph Neural Networks (GNNs) often suffer from degree bias in node classification tasks, where prediction performance varies across nodes with different degrees. Several approaches, which adopt Graph Contrastive Learning (GCL), have been proposed to mitigate this bias. However, the limited number of positive pairs and the equal weighting of all positives and negatives in GCL still lead to low-degree nodes acquiring insufficient and noisy information. This paper proposes the Hardness Adaptive Reweighted (HAR) contrastive loss to mitigate degree bias. It adds more positive pairs by leveraging node labels and adaptively weights positive and negative pairs based on their learning hardness. In addition, we develop an experimental framework named SHARP to extend HAR to a broader range of scenarios. Both our theoretical analysis and experiments validate the effectiveness of SHARP. The experimental results across four datasets show that SHARP achieves better performance against baselines at both global and degree levels. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 16:28:12 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Hu",
"Jingyu",
""
],
[
"Bo",
"Hongbo",
""
],
[
"Hong",
"Jun",
""
],
[
"Liu",
"Xiaowei",
""
],
[
"Liu",
"Weiru",
""
]
] |
2506.05215 | Zhan Yu | Zhan Yu | Learning Theory of Decentralized Robust Kernel-Based Learning Algorithm | null | null | null | null | cs.LG | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | We propose a new decentralized robust kernel-based learning algorithm within the framework of reproducing kernel Hilbert space (RKHS) by utilizing a networked system that can be represented as a connected graph. The robust loss function $\mathcal{L}_\sigma$ induced by a windowing function $W$ and a robustness scaling parameter $\sigma>0$, can encompass a broad spectrum of robust losses. Consequently, the proposed algorithm effectively provides a unified decentralized learning framework for robust regression, which fundamentally differs from the existing distributed robust kernel learning schemes, all of which are divide-and-conquer based. We rigorously establish the learning theory and offer a comprehensive convergence analysis for the algorithm. We show each local robust estimator generated from the decentralized algorithm can be utilized to approximate the regression function. Based on kernel-based integral operator techniques, we derive general high confidence convergence bounds for each local approximating sequence in terms of the mean square distance, RKHS norm, and generalization error, respectively. Moreover, we provide rigorous selection rules for local sample size and show that, under properly selected step size and scaling parameter $\sigma$, the decentralized robust algorithm can achieve optimal learning rates (up to logarithmic factors) in both norms. The parameter $\sigma$ is shown to be essential for enhancing robustness while also ensuring favorable convergence behavior. The intrinsic connection among decentralization, sample selection, robustness of the algorithm, and its convergence is clearly reflected. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 16:30:05 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Yu",
"Zhan",
""
]
] |
2506.05216 | Akshay Seshadri | Tyler Chen, Akshay Seshadri, Mattia J. Villani, Pradeep Niroula, Shouvanik Chakrabarti, Archan Ray, Pranav Deshpande, Romina Yalovetzky, Marco Pistoia, Niraj Kumar | A Unified Framework for Provably Efficient Algorithms to Estimate Shapley Values | 44 pages, 7 figures, 7 tables | null | null | null | cs.LG cs.DS quant-ph | http://creativecommons.org/licenses/by-nc-nd/4.0/ | Shapley values have emerged as a critical tool for explaining which features impact the decisions made by machine learning models. However, computing exact Shapley values is difficult, generally requiring an exponential (in the feature dimension) number of model evaluations. To address this, many model-agnostic randomized estimators have been developed, the most influential and widely used being the KernelSHAP method (Lundberg & Lee, 2017). While related estimators such as unbiased KernelSHAP (Covert & Lee, 2021) and LeverageSHAP (Musco & Witter, 2025) are known to satisfy theoretical guarantees, bounds for KernelSHAP have remained elusive. We describe a broad and unified framework that encompasses KernelSHAP and related estimators constructed using both with and without replacement sampling strategies. We then prove strong non-asymptotic theoretical guarantees that apply to all estimators from our framework. This provides, to the best of our knowledge, the first theoretical guarantees for KernelSHAP and sheds further light on tradeoffs between existing estimators. Through comprehensive benchmarking on small and medium dimensional datasets for Decision-Tree models, we validate our approach against exact Shapley values, consistently achieving low mean squared error with modest sample sizes. Furthermore, we make specific implementation improvements to enable scalability of our methods to high-dimensional datasets. Our methods, tested on datasets such MNIST and CIFAR10, provide consistently better results compared to the KernelSHAP library. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 16:30:53 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Chen",
"Tyler",
""
],
[
"Seshadri",
"Akshay",
""
],
[
"Villani",
"Mattia J.",
""
],
[
"Niroula",
"Pradeep",
""
],
[
"Chakrabarti",
"Shouvanik",
""
],
[
"Ray",
"Archan",
""
],
[
"Deshpande",
"Pranav",
""
],
[
"Yalovetzky",
"Romina",
""
],
[
"Pistoia",
"Marco",
""
],
[
"Kumar",
"Niraj",
""
]
] |
2506.05217 | Wenhao Hu | Wenhao Hu, Xuexiang Wen, Xi Li, Gaoang Wang | DSG-World: Learning a 3D Gaussian World Model from Dual State Videos | null | null | null | null | cs.CV | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Building an efficient and physically consistent world model from limited observations is a long standing challenge in vision and robotics. Many existing world modeling pipelines are based on implicit generative models, which are hard to train and often lack 3D or physical consistency. On the other hand, explicit 3D methods built from a single state often require multi-stage processing-such as segmentation, background completion, and inpainting-due to occlusions. To address this, we leverage two perturbed observations of the same scene under different object configurations. These dual states offer complementary visibility, alleviating occlusion issues during state transitions and enabling more stable and complete reconstruction. In this paper, we present DSG-World, a novel end-to-end framework that explicitly constructs a 3D Gaussian World model from Dual State observations. Our approach builds dual segmentation-aware Gaussian fields and enforces bidirectional photometric and semantic consistency. We further introduce a pseudo intermediate state for symmetric alignment and design collaborative co-pruning trategies to refine geometric completeness. DSG-World enables efficient real-to-simulation transfer purely in the explicit Gaussian representation space, supporting high-fidelity rendering and object-level scene manipulation without relying on dense observations or multi-stage pipelines. Extensive experiments demonstrate strong generalization to novel views and scene states, highlighting the effectiveness of our approach for real-world 3D reconstruction and simulation. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 16:33:32 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Hu",
"Wenhao",
""
],
[
"Wen",
"Xuexiang",
""
],
[
"Li",
"Xi",
""
],
[
"Wang",
"Gaoang",
""
]
] |
2506.05218 | Yuliang Liu | Zhang Li, Yuliang Liu, Qiang Liu, Zhiyin Ma, Ziyang Zhang, Shuo Zhang, Zidun Guo, Jiarui Zhang, Xinyu Wang, Xiang Bai | MonkeyOCR: Document Parsing with a Structure-Recognition-Relation Triplet Paradigm | null | null | null | null | cs.CV | http://creativecommons.org/licenses/by-nc-sa/4.0/ | We introduce MonkeyOCR, a vision-language model for document parsing that advances the state of the art by leveraging a Structure-Recognition-Relation (SRR) triplet paradigm. This design simplifies what would otherwise be a complex multi-tool pipeline (as in MinerU's modular approach) and avoids the inefficiencies of processing full pages with giant end-to-end models (e.g., large multimodal LLMs like Qwen-VL). In SRR, document parsing is abstracted into three fundamental questions - "Where is it?" (structure), "What is it?" (recognition), and "How is it organized?" (relation) - corresponding to layout analysis, content identification, and logical ordering. This focused decomposition balances accuracy and speed: it enables efficient, scalable processing without sacrificing precision. To train and evaluate this approach, we introduce the MonkeyDoc (the most comprehensive document parsing dataset to date), with 3.9 million instances spanning over ten document types in both Chinese and English. Experiments show that MonkeyOCR outperforms MinerU by an average of 5.1%, with particularly notable improvements on challenging content such as formulas (+15.0%) and tables (+8.6%). Remarkably, our 3B-parameter model surpasses much larger and top-performing models, including Qwen2.5-VL (72B) and Gemini 2.5 Pro, achieving state-of-the-art average performance on English document parsing tasks. In addition, MonkeyOCR processes multi-page documents significantly faster (0.84 pages per second compared to 0.65 for MinerU and 0.12 for Qwen2.5-VL-7B). The 3B model can be efficiently deployed for inference on a single NVIDIA 3090 GPU. Code and models will be released at https://github.com/Yuliang-Liu/MonkeyOCR. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 16:34:57 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Li",
"Zhang",
""
],
[
"Liu",
"Yuliang",
""
],
[
"Liu",
"Qiang",
""
],
[
"Ma",
"Zhiyin",
""
],
[
"Zhang",
"Ziyang",
""
],
[
"Zhang",
"Shuo",
""
],
[
"Guo",
"Zidun",
""
],
[
"Zhang",
"Jiarui",
""
],
[
"Wang",
"Xinyu",
""
],
[
"Bai",
"Xiang",
""
]
] |
2506.05219 | EPTCS | Cynthia Kop (Radboud Universiteit Nijmegen), Helida Salles Santos (Universidade Federal do Rio Grande) | Proceedings of the 19th International Workshop on Logical and Semantic Frameworks, with Applications | null | EPTCS 421, 2025 | 10.4204/EPTCS.421 | null | cs.LO cs.SC | http://creativecommons.org/licenses/by/4.0/ | This volume contains the post-proceedings of the 19th LSFA, which was held in Goi\^ania, the capital of Goi\'as state in Brazil, from September 18 to September 20, 2024.
Logical and semantic frameworks are formal languages used to represent logics, languages and systems. These frameworks provide foundations for the formal specification of systems and programming languages, supporting tool development and reasoning.
The aim of this series is bringing together theoreticians and practitioners to promote new techniques and results, from the theoretical side, and feedback on the implementation and use of such techniques and results, from the practical side. LSFA includes areas such as proof and type theory, equational deduction and rewriting systems, automated reasoning and concurrency theory.
| [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 16:35:22 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Kop",
"Cynthia",
"",
"Radboud Universiteit Nijmegen"
],
[
"Santos",
"Helida Salles",
"",
"Universidade Federal do Rio Grande"
]
] |
2506.05220 | Arnaud Huber | A. Huber, D. Raffestin, F. Hannachi, M. Huault, J. Jouve, A. Maitrallain, P. Nicola\"i, B. Zielbauer, and M. Tarisien | Design of an Active Thomson Parabola for the detection of ions accelerated by laser: Numerical simulations and characterization of different solutions | null | null | null | null | physics.ins-det physics.acc-ph physics.plasm-ph | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | This article describes the development of an active Thomson Parabola ion spectrometer designed to measure the energy spectra of different multi-MeV ion species generated in laser-plasmas interactions. To do so, GEANT4 optical simulations were carried out in order to design and build a spectrometer based on scintillator detectors capable of operating at rates comparable with the highest achievable at any existing or upcoming high intensity laser facilities. Details on the different configurations simulated and their characterization are discussed. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 16:37:17 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Huber",
"A.",
""
],
[
"Raffestin",
"D.",
""
],
[
"Hannachi",
"F.",
""
],
[
"Huault",
"M.",
""
],
[
"Jouve",
"J.",
""
],
[
"Maitrallain",
"A.",
""
],
[
"Nicolaï",
"P.",
""
],
[
"Zielbauer",
"B.",
""
],
[
"Tarisien",
"M.",
""
]
] |
2506.05221 | Jianghao Wu | Jianghao Wu, Yicheng Wu, Yutong Xie, Wenjia Bai, You Zhang, Feilong Tang, Yulong Li, Yasmeen George, Imran Razzak | SAM-aware Test-time Adaptation for Universal Medical Image Segmentation | 10 pages, 4 figures | null | null | null | cs.CV | http://creativecommons.org/licenses/by/4.0/ | Universal medical image segmentation using the Segment Anything Model (SAM) remains challenging due to its limited adaptability to medical domains. Existing adaptations, such as MedSAM, enhance SAM's performance in medical imaging but at the cost of reduced generalization to unseen data. Therefore, in this paper, we propose SAM-aware Test-Time Adaptation (SAM-TTA), a fundamentally different pipeline that preserves the generalization of SAM while improving its segmentation performance in medical imaging via a test-time framework. SAM-TTA tackles two key challenges: (1) input-level discrepancies caused by differences in image acquisition between natural and medical images and (2) semantic-level discrepancies due to fundamental differences in object definition between natural and medical domains (e.g., clear boundaries vs. ambiguous structures). Specifically, our SAM-TTA framework comprises (1) Self-adaptive Bezier Curve-based Transformation (SBCT), which adaptively converts single-channel medical images into three-channel SAM-compatible inputs while maintaining structural integrity, to mitigate the input gap between medical and natural images, and (2) Dual-scale Uncertainty-driven Mean Teacher adaptation (DUMT), which employs consistency learning to align SAM's internal representations to medical semantics, enabling efficient adaptation without auxiliary supervision or expensive retraining. Extensive experiments on five public datasets demonstrate that our SAM-TTA outperforms existing TTA approaches and even surpasses fully fine-tuned models such as MedSAM in certain scenarios, establishing a new paradigm for universal medical image segmentation. Code can be found at https://github.com/JianghaoWu/SAM-TTA. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 16:38:16 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Wu",
"Jianghao",
""
],
[
"Wu",
"Yicheng",
""
],
[
"Xie",
"Yutong",
""
],
[
"Bai",
"Wenjia",
""
],
[
"Zhang",
"You",
""
],
[
"Tang",
"Feilong",
""
],
[
"Li",
"Yulong",
""
],
[
"George",
"Yasmeen",
""
],
[
"Razzak",
"Imran",
""
]
] |
2506.05222 | Pappu Acharya | Pappu Acharya, Romain Mari | Modelling the evolution of flow-induced anisotropy of concentrated suspensions | 4 pages, 2 figures, scheduled to be published in "EPJ Web of Conferences'' (Powders and Grains 2025) | null | null | null | cond-mat.soft | http://creativecommons.org/licenses/by/4.0/ | Suspensions, which exhibit complex behaviors such as shear thickening, thinning, and jamming, are prevalent in nature and industry. However, predicting the mechanical properties of concentrated suspensions, in both steady state and the transient regime, remains a significant challenge, impacting product quality and process efficiency. In this study, we focus on developing a robust theoretical framework to explain how flow history governs the anisotropy of mechanical responses in suspensions of hard particles under unsteady flow conditions. Our starting point is the Gillissen-Wilson constitutive model, which we confront to DEM simulation data of the micro-structure during steady shear, and shear rotations where the shear axis is rotated by a specific angle around the flow gradient direction. We introduce a simple modification to the Gillissen-Wilson model which leads to a model with higher predictive power in steady state and during shear rotations. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 16:38:17 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Acharya",
"Pappu",
""
],
[
"Mari",
"Romain",
""
]
] |
2506.05223 | Guy Vosco | Guy Vosco and Sivan Refaely-Abramson | Exciton-Exciton Annihilation Mediated by Many-Body Coulomb and Phonon Interactions: An Ab Initio Study | null | null | null | null | cond-mat.mtrl-sci | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Exciton-exciton annihilation (EEA), in which two excitons interact to generate high-energy excitations, is an important non-radiative channel in light-induced excited-state relaxation. When efficient, this process offers an alternative route to exciton emission, potentially allowing extended energetically excited particles' lifetime and coherence. These properties are significant in designing and understanding materials-based quantum devices, particularly for low-dimensional semiconductors. Here, we present a first-principles framework to compute EEA mechanisms and rates using many-body perturbation theory within the GW and Bethe-Salpeter Equation (GW-BSE) formalism. Our method explicitly accounts for Coulomb-driven and phonon-assisted exciton-exciton scattering by explicitly evaluating the interaction channels between the constituent electrons and holes composing the BSE excitons. We apply this framework to monolayer WSe$_2$ and explore the $A$, $B$ excitation manifolds, finding picosecond-scale annihilation between bright and dark states, cross valleys, and cross peak manifolds. These channels become allowed due to scattering into free electron-hole pairs across the Brillouin zone. Our results supply new insights into non-radiative exciton relaxation mechanisms in two-dimensional materials, providing a predictive and general tool for modeling these interactions in excitonic materials. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 16:38:54 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Vosco",
"Guy",
""
],
[
"Refaely-Abramson",
"Sivan",
""
]
] |
2506.05224 | Jose L. Goity | D. Jayakodige and J. L. Goity | Pion-Nucleon Scattering in Baryon Chiral Perturbation Theory combined with the ${ 1/N_c}$ Expansion | 54 pages, 11 figures | null | null | JLAB-THY-25-4343 | hep-ph nucl-th | http://creativecommons.org/licenses/by-nc-nd/4.0/ | This work implements the combined BChPT and $1/N_c$ expansions for pion-nucleon elastic scattering. The effective theory is based on the baryon sector dynamical spin-flavor $SU(4)$ symmetry emergent in the large $N_c$ limit, whose breaking in controlled by the $1/N_c$ expansion. The non-commutativity of the chiral and $1/N_c$ expansions in unitarity corrections (loops) requires a linking of both expansions. As it was shown in the case of baryon masses and currents, the natural linking is the $\xi$-expansion, in which ${\cal{O}}(p)={\cal{O}}(1/N_c)={\cal{O}}(\xi)$. The spin-flavor symmetry requires that the ground state baryons span an $SU(4)$ symmetric irreducible representation which implies that in particular $N$ and $\Delta$ are active degrees of freedom in the effective theory.
The scattering amplitude is expanded to the next-to-next to leading order in the $\xi$ expansion, corresponding to the one-loop contributions with the LO Lagrangian. The results are given for generic $N_c$ in order to demonstrate the consistency of the framework. The spin-flavor symmetry plays a central role in maintaining the consistency of the effective theory with respect to the $1/N_c$ expansion. This consistency manifests itself in an improvement in the convergence of the low energy expansion with respect to the case of the ordinary BChPT without an explicit dynamical $\Delta$, which is known to be inconsistent the constraints of $N_c$ scaling.
Fits to the $\pi N\to \pi N$ S, P and D partial wave amplitudes from the SAID data base are finally used to test the framework and to determine the energy range of its applicability. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 16:39:39 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Jayakodige",
"D.",
""
],
[
"Goity",
"J. L.",
""
]
] |
2506.05225 | Jack Collison | Harold D. Chiang, Jack Collison, Lorenzo Magnolfi, and Christopher Sullivan | Enhancing the Merger Simulation Toolkit with ML/AI | 18 pages, 4 figures | null | null | null | econ.EM | http://creativecommons.org/licenses/by/4.0/ | This paper develops a flexible approach to predict the price effects of horizontal mergers using ML/AI methods. While standard merger simulation techniques rely on restrictive assumptions about firm conduct, we propose a data-driven framework that relaxes these constraints when rich market data are available. We develop and identify a flexible nonparametric model of supply that nests a broad range of conduct models and cost functions. To overcome the curse of dimensionality, we adapt the Variational Method of Moments (VMM) (Bennett and Kallus, 2023) to estimate the model, allowing for various forms of strategic interaction. Monte Carlo simulations show that our method significantly outperforms an array of misspecified models and rivals the performance of the true model, both in predictive performance and counterfactual merger simulations. As a way to interpret the economics of the estimated function, we simulate pass-through and reveal that the model learns markup and cost functions that imply approximately correct pass-through behavior. Applied to the American Airlines-US Airways merger, our method produces more accurate post-merger price predictions than traditional approaches. The results demonstrate the potential for machine learning techniques to enhance merger analysis while maintaining economic structure. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 16:40:00 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Chiang",
"Harold D.",
""
],
[
"Collison",
"Jack",
""
],
[
"Magnolfi",
"Lorenzo",
""
],
[
"Sullivan",
"Christopher",
""
]
] |
2506.05226 | Mohammed Almutairi K | Mohammed Almutairi and Diego G\'omez-Zar\'a | Towards Effective Multidisciplinary Health and HCI Teams based on AI Framework | 4 pages, 1 figure, Presented at CHI24, Workshop on Conducting Research at the Intersection of HCI and Health Building and Supporting Teams with Diverse Expertise to Increase Public Health Impact, May 11 to 16, 2024, Honolulu, HI, USA | null | null | null | cs.HC | http://creativecommons.org/licenses/by/4.0/ | As a Ph.D. student with a diverse background in both public and private sectors, I have encountered numerous challenges in cross-disciplinary and multi-stakeholder team projects. My research on developing team compositions that involve multidisciplinary members from fields including education, academia, and health. Along with my advisor, we are focused on exploring how HCI can help individuals assemble more effective teams. This effort involves developing socio-technical systems that guide and inform individuals of the potential teams that they can assemble. We employ state-of-the-art algorithms that prioritize inclusion among team members from diverse areas of expertise and familiarity between the team members. Our goal for attending this workshop is to engage in meaningful dialogues with scholars and researchers, leveraging these interactions to refine our approach to building an AI-driven team composition system to foster effective, interdisciplinary collaboration in health-focused HCI research. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 16:40:41 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Almutairi",
"Mohammed",
""
],
[
"Gómez-Zará",
"Diego",
""
]
] |
2506.05227 | Adam Wiemerslage | Adam Wiemerslage and Katharina von der Wense | Improving Low-Resource Morphological Inflection via Self-Supervised Objectives | ACL 2025 main | null | null | null | cs.CL | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Self-supervised objectives have driven major advances in NLP by leveraging large-scale unlabeled data, but such resources are scarce for many of the world's languages. Surprisingly, they have not been explored much for character-level tasks, where smaller amounts of data have the potential to be beneficial. We investigate the effectiveness of self-supervised auxiliary tasks for morphological inflection -- a character-level task highly relevant for language documentation -- in extremely low-resource settings, training encoder-decoder transformers for 19 languages and 13 auxiliary objectives. Autoencoding yields the best performance when unlabeled data is very limited, while character masked language modeling (CMLM) becomes more effective as data availability increases. Though objectives with stronger inductive biases influence model predictions intuitively, they rarely outperform standard CMLM. However, sampling masks based on known morpheme boundaries consistently improves performance, highlighting a promising direction for low-resource morphological modeling. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 16:42:45 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Wiemerslage",
"Adam",
""
],
[
"von der Wense",
"Katharina",
""
]
] |
2506.05228 | XiaoHui Yang | Xiaohui Yang, Lei Yang, Xinhui Lu, Yu Guo | Spherical Phase Metalenses: Intrinsic Suppression of Spherical Aberration via Equiphase Surface Modulation | null | null | null | null | physics.optics | http://creativecommons.org/licenses/by/4.0/ | Recent progress in large-scale metasurfaces requires phase profiles beyond traditional hyperbolic designs. We show hyperbolic phase distributions cause spherical aberration from mismatched light propagation geometry and unrealistic phase assumptions. By analyzing metalens fundamentals via isophase surfaces, we develop a spherical phase profile based on spherical wavefront theory. This method prevents spherical aberration, essential for wide-aperture metalenses. Simulations prove superior focusing: spherical phase reduces FWHM by 7.3% and increases peak intensity by 20.4% versus hyperbolic designs at 31.46 micron radius. Spherical phase maintains consistent focusing across radii, while hyperbolic phase shows strong correlation (R squared = 0.95) with aberration. We also propose a normal vector tracing metric to measure design aberrations. This work establishes a scalable framework for diffraction-limited metalenses. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 16:43:12 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Yang",
"Xiaohui",
""
],
[
"Yang",
"Lei",
""
],
[
"Lu",
"Xinhui",
""
],
[
"Guo",
"Yu",
""
]
] |
2506.05229 | Danil Sivtsov | Danil Sivtsov, Ivan Rodkin, Gleb Kuzmin, Yuri Kuratov, Ivan Oseledets | Diagonal Batching Unlocks Parallelism in Recurrent Memory Transformers for Long Contexts | null | null | null | null | cs.LG cs.CL | http://creativecommons.org/licenses/by/4.0/ | Transformer models struggle with long-context inference due to their quadratic time and linear memory complexity. Recurrent Memory Transformers (RMTs) offer a solution by reducing the asymptotic cost to linear time and constant memory usage. However, their memory update mechanism leads to sequential execution, causing a performance bottleneck.
We introduce Diagonal Batching, a scheduling scheme that unlocks parallelism across segments in RMTs while preserving exact recurrence. This approach eliminates the sequential constraint, enabling efficient GPU inference even for single long-context inputs without complex batching and pipelining techniques. Because the technique is purely a run-time computation reordering, existing RMT models adopt it with no retraining.
Applied to a LLaMA-1B ARMT model, Diagonal Batching yields a 3.3x speedup over standard full-attention LLaMA-1B and a 1.8x speedup over the sequential RMT implementation on 131,072-token sequences. By removing sequential bottleneck, Diagonal Batching reduces inference cost and latency, thereby strengthening RMTs as a practical solution for real-world, long-context applications. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 16:43:48 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Sivtsov",
"Danil",
""
],
[
"Rodkin",
"Ivan",
""
],
[
"Kuzmin",
"Gleb",
""
],
[
"Kuratov",
"Yuri",
""
],
[
"Oseledets",
"Ivan",
""
]
] |
2506.05230 | Tiago Ant\~ao | Tiago V. C. Ant\~ao, Yitao Sun, Adolfo O. Fumega, Jose L. Lado | Tensor network method for real-space topology in quasicrystal Chern mosaics | 7 pages, 3 figures | null | null | null | cond-mat.str-el | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Computing topological invariants in two-dimensional quasicrystals and super-moire matter is an exceptional open challenge, due to the absence of translational symmetry and the colossal number of sites inherent to these systems. Here, we establish a method to compute local topological invariants of exceptionally large systems using tensor networks, enabling the computation of invariants for Hamiltonians with hundreds of millions of sites, several orders of magnitude above the capabilities of conventional methodologies. Our approach leverages a tensor-network representation of the density matrix using a Chebyshev tensor network algorithm, enabling large-scale calculations of topological markers in quasicrystalline and moire systems. We demonstrate our methodology with two-dimensional quasicrystals featuring 8-fold and 10-fold rotational symmetries and mosaics of Chern phases. Our work establishes a powerful method to compute topological phases in exceptionally large-scale topological systems, establishing the required tool to rationalize generic super-moire and quasicrystalline topological matter. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 16:46:00 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Antão",
"Tiago V. C.",
""
],
[
"Sun",
"Yitao",
""
],
[
"Fumega",
"Adolfo O.",
""
],
[
"Lado",
"Jose L.",
""
]
] |
2506.05231 | Wenlin Chen | Severi Rissanen, RuiKang OuYang, Jiajun He, Wenlin Chen, Markus Heinonen, Arno Solin, Jos\'e Miguel Hern\'andez-Lobato | Progressive Tempering Sampler with Diffusion | Accepted for publication at ICML 2025 | null | null | null | cs.LG stat.ML | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Recent research has focused on designing neural samplers that amortize the process of sampling from unnormalized densities. However, despite significant advancements, they still fall short of the state-of-the-art MCMC approach, Parallel Tempering (PT), when it comes to the efficiency of target evaluations. On the other hand, unlike a well-trained neural sampler, PT yields only dependent samples and needs to be rerun -- at considerable computational cost -- whenever new samples are required. To address these weaknesses, we propose the Progressive Tempering Sampler with Diffusion (PTSD), which trains diffusion models sequentially across temperatures, leveraging the advantages of PT to improve the training of neural samplers. We also introduce a novel method to combine high-temperature diffusion models to generate approximate lower-temperature samples, which are minimally refined using MCMC and used to train the next diffusion model. PTSD enables efficient reuse of sample information across temperature levels while generating well-mixed, uncorrelated samples. Our method significantly improves target evaluation efficiency, outperforming diffusion-based neural samplers. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 16:46:04 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Rissanen",
"Severi",
""
],
[
"OuYang",
"RuiKang",
""
],
[
"He",
"Jiajun",
""
],
[
"Chen",
"Wenlin",
""
],
[
"Heinonen",
"Markus",
""
],
[
"Solin",
"Arno",
""
],
[
"Hernández-Lobato",
"José Miguel",
""
]
] |
2506.05232 | Suwei Yang | Suwei Yang, Yong Lai, Kuldeep S. Meel | On Top-Down Pseudo-Boolean Model Counting | To appear in SAT 2025 | null | null | null | cs.LO | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Pseudo-Boolean model counting involves computing the number of satisfying assignments of a given pseudo-Boolean (PB) formula. In recent years, PB model counting has seen increased interest partly owing to the succinctness of PB formulas over typical propositional Boolean formulas in conjunctive normal form (CNF) at describing problem constraints. In particular, the research community has developed tools to tackle exact PB model counting. These recently developed counters follow one of the two existing major designs for model counters, namely the bottom-up model counter design. A natural question would be whether the other major design, the top-down model counter paradigm, would be effective at PB model counting, especially when the top-down design offered superior performance in CNF model counting literature.
In this work, we investigate the aforementioned top-down design for PB model counting and introduce the first exact top-down PB model counter, PBMC. PBMC is a top-down search-based counter for PB formulas, with a new variable decision heuristic that considers variable coefficients. Through our evaluations, we highlight the superior performance of PBMC at PB model counting compared to the existing state-of-the-art counters PBCount, PBCounter, and Ganak. In particular, PBMC could count for 1849 instances while the next-best competing method, PBCount, could only count for 1773 instances, demonstrating the potential of a top-down PB counter design. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 16:49:28 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Yang",
"Suwei",
""
],
[
"Lai",
"Yong",
""
],
[
"Meel",
"Kuldeep S.",
""
]
] |
2506.05233 | Nino Scherrer | Johannes von Oswald, Nino Scherrer, Seijin Kobayashi, Luca Versari, Songlin Yang, Maximilian Schlegel, Kaitlin Maile, Yanick Schimpf, Oliver Sieberling, Alexander Meulemans, Rif A. Saurous, Guillaume Lajoie, Charlotte Frenkel, Razvan Pascanu, Blaise Ag\"uera y Arcas, Jo\~ao Sacramento | MesaNet: Sequence Modeling by Locally Optimal Test-Time Training | null | null | null | null | cs.LG cs.AI cs.CL | http://creativecommons.org/licenses/by/4.0/ | Sequence modeling is currently dominated by causal transformer architectures that use softmax self-attention. Although widely adopted, transformers require scaling memory and compute linearly during inference. A recent stream of work linearized the softmax operation, resulting in powerful recurrent neural network (RNN) models with constant memory and compute costs such as DeltaNet, Mamba or xLSTM. These models can be unified by noting that their recurrent layer dynamics can all be derived from an in-context regression objective, approximately optimized through an online learning rule. Here, we join this line of work and introduce a numerically stable, chunkwise parallelizable version of the recently proposed Mesa layer (von Oswald et al., 2024), and study it in language modeling at the billion-parameter scale. This layer again stems from an in-context loss, but which is now minimized to optimality at every time point using a fast conjugate gradient solver. Through an extensive suite of experiments, we show that optimal test-time training enables reaching lower language modeling perplexity and higher downstream benchmark performance than previous RNNs, especially on tasks requiring long context understanding. This performance gain comes at the cost of additional flops spent during inference time. Our results are therefore intriguingly related to recent trends of increasing test-time compute to improve performance -- here by spending compute to solve sequential optimization problems within the neural network itself. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 16:50:23 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"von Oswald",
"Johannes",
""
],
[
"Scherrer",
"Nino",
""
],
[
"Kobayashi",
"Seijin",
""
],
[
"Versari",
"Luca",
""
],
[
"Yang",
"Songlin",
""
],
[
"Schlegel",
"Maximilian",
""
],
[
"Maile",
"Kaitlin",
""
],
[
"Schimpf",
"Yanick",
""
],
[
"Sieberling",
"Oliver",
""
],
[
"Meulemans",
"Alexander",
""
],
[
"Saurous",
"Rif A.",
""
],
[
"Lajoie",
"Guillaume",
""
],
[
"Frenkel",
"Charlotte",
""
],
[
"Pascanu",
"Razvan",
""
],
[
"Arcas",
"Blaise Agüera y",
""
],
[
"Sacramento",
"João",
""
]
] |
2506.05234 | Tien-Tien Yu | Tien-Tien Yu | 2024 TASI Lectures: A Dark Matter Primer | 30 pages, 13 figures. Comments welcomed | null | null | null | hep-ph astro-ph.CO | http://creativecommons.org/licenses/by/4.0/ | These notes are based on a sequence of 4 lectures delivered at the 2024 Theoretical Advanced Study Institute (TASI) and at the Universit\`a degli Studi di Padova. They are intended for graduate students at the early stages of their study of dark matter with some prior exposure to cosmology and quantum field theory. The primary aim is to offer an accessible introduction to dark matter and to lay the groundwork for exploring its phenomenology. These lectures are not intended to serve as a comprehensive review. We begin by motivating the study of dark matter through a discussion of the empirical evidence and the constraints it places on dark matter properties. This is followed by an overview of several canonical mechanisms for the production of cosmological dark matter, and a concluding section that surveys current experimental and observational efforts to detect it with a focus on direct detection. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 16:51:16 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Yu",
"Tien-Tien",
""
]
] |
2506.05235 | Ratna Kandala | Ratna Kandala, Prakash Mondal | Towards a Unified System of Representation for Continuity and Discontinuity in Natural Language | null | null | null | null | cs.CL | http://creativecommons.org/licenses/by-nc-sa/4.0/ | Syntactic discontinuity is a grammatical phenomenon in which a constituent is split into more than one part because of the insertion of an element which is not part of the constituent. This is observed in many languages across the world such as Turkish, Russian, Japanese, Warlpiri, Navajo, Hopi, Dyirbal, Yidiny etc. Different formalisms/frameworks in current linguistic theory approach the problem of discontinuous structures in different ways. Each framework/formalism has widely been viewed as an independent and non-converging system of analysis. In this paper, we propose a unified system of representation for both continuity and discontinuity in structures of natural languages by taking into account three formalisms, in particular, Phrase Structure Grammar (PSG) for its widely used notion of constituency, Dependency Grammar (DG) for its head-dependent relations, and Categorial Grammar (CG) for its focus on functor-argument relations. We attempt to show that discontinuous expressions as well as continuous structures can be analysed through a unified mathematical derivation incorporating the representations of linguistic structure in these three grammar formalisms. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 16:54:41 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Kandala",
"Ratna",
""
],
[
"Mondal",
"Prakash",
""
]
] |
2506.05236 | Maxime Toquebiau | Maxime Toquebiau, Jae-Yun Jun, Fa\"iz Benamar, Nicolas Bredeche | Towards Language-Augmented Multi-Agent Deep Reinforcement Learning | null | null | null | null | cs.MA | http://creativecommons.org/licenses/by/4.0/ | Communication is a fundamental aspect of coordinated behavior in multi-agent reinforcement learning. Yet, most prior works in this field have focused on emergent communication protocols developed from scratch, often resulting in inefficient or non-interpretable systems. Inspired by the role of language in natural intelligence, we investigate how grounding agents in a human-defined language can improve learning and coordination of multiple embodied agents. We propose a framework in which agents are trained not only to act but also to produce and interpret natural language descriptions of their observations. This language-augmented learning serves a dual role: enabling explicit communication between agents and guiding representation learning. We demonstrate that agents trained with our method outperform traditional emergent communication baselines across various tasks. Our analysis reveals that language grounding leads to more informative internal representations, better generalization to new partners, and improved capability for human-agent interaction. These findings demonstrate the effectiveness of integrating structured language into multi-agent learning and open avenues for more interpretable and capable multi-agent systems. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 16:55:52 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Toquebiau",
"Maxime",
""
],
[
"Jun",
"Jae-Yun",
""
],
[
"Benamar",
"Faïz",
""
],
[
"Bredeche",
"Nicolas",
""
]
] |
2506.05237 | Victoria Menescal Tupper Palhares | Victoria Palhares, Sueda Taner, and Christoph Studer | CSI2Vec: Towards a Universal CSI Feature Representation for Positioning and Channel Charting | Submitted to a journal | null | null | null | cs.IT eess.SP math.IT | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Natural language processing techniques, such as Word2Vec, have demonstrated exceptional capabilities in capturing semantic and syntactic relationships of text through vector embeddings. Inspired by this technique, we propose CSI2Vec, a self-supervised framework for generating universal and robust channel state information (CSI) representations tailored to CSI-based positioning (POS) and channel charting (CC). CSI2Vec learns compact vector embeddings across various wireless scenarios, capturing spatial relationships between user equipment positions without relying on CSI reconstruction or ground-truth position information. We implement CSI2Vec as a neural network that is trained across various deployment setups (i.e., the spatial arrangement of radio equipment and scatterers) and radio setups (RSs) (i.e., the specific hardware used), ensuring robustness to aspects such as differences in the environment, the number of used antennas, or allocated set of subcarriers. CSI2Vec abstracts the RS by generating compact vector embeddings that capture essential spatial information, avoiding the need for full CSI transmission or reconstruction while also reducing complexity and improving processing efficiency of downstream tasks. Simulations with ray-tracing and real-world CSI datasets demonstrate CSI2Vec's effectiveness in maintaining excellent POS and CC performance while reducing computational demands and storage. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 16:56:41 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Palhares",
"Victoria",
""
],
[
"Taner",
"Sueda",
""
],
[
"Studer",
"Christoph",
""
]
] |
2506.05238 | Victor Saunier | Jordan Levin, Guglielmo Nocera, Victor Saunier | A model for the assembly map of bordism-invariant functors | 42 pages; comments welcome! | null | null | null | math.KT math.AT math.CT math.GT | http://creativecommons.org/licenses/by-nc-sa/4.0/ | We study oplax colimits of stable categories, of hermitian categories and of Poincar\'e categories in nice cases. This allows us to produce a categorical model of the assembly map of a bordism-invariant functor of Poincar\'e categories which is also a Verdier projection, whose kernel we explicitly describe. As a direct application, we generalize the Shaneson splitting for bordism-invariant functors of Poincar\'e categories proved by Calm\`es-Dotto-Harpaz-Hebestreit-Land-Moi-Nardin-Nikolaus-Steimle to allow for twists. We also show our methods can tackle their general twisted Shaneson splitting of Poincar\'e-Verdier localizing invariants which specifies to a twisted Bass-Heller-Swan decomposition for the underlying stable categories, generalizing part of recent work of Kirstein-Kremer. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 16:57:49 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Levin",
"Jordan",
""
],
[
"Nocera",
"Guglielmo",
""
],
[
"Saunier",
"Victor",
""
]
] |
2506.05239 | Val\'erie Costa | Val\'erie Costa, Thomas Fel, Ekdeep Singh Lubana, Bahareh Tolooshams, Demba Ba | Evaluating Sparse Autoencoders: From Shallow Design to Matching Pursuit | Complementary work to arXiv:2506.03093 | null | null | null | cs.LG | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Sparse autoencoders (SAEs) have recently become central tools for interpretability, leveraging dictionary learning principles to extract sparse, interpretable features from neural representations whose underlying structure is typically unknown. This paper evaluates SAEs in a controlled setting using MNIST, which reveals that current shallow architectures implicitly rely on a quasi-orthogonality assumption that limits the ability to extract correlated features. To move beyond this, we introduce a multi-iteration SAE by unrolling Matching Pursuit (MP-SAE), enabling the residual-guided extraction of correlated features that arise in hierarchical settings such as handwritten digit generation while guaranteeing monotonic improvement of the reconstruction as more atoms are selected. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 16:57:58 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Costa",
"Valérie",
""
],
[
"Fel",
"Thomas",
""
],
[
"Lubana",
"Ekdeep Singh",
""
],
[
"Tolooshams",
"Bahareh",
""
],
[
"Ba",
"Demba",
""
]
] |
2506.05240 | Yizhuo Li | Yizhuo Li, Yuying Ge, Yixiao Ge, Ying Shan, Ping Luo | Aligning Latent Spaces with Flow Priors | null | null | null | null | cs.LG cs.CV | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | This paper presents a novel framework for aligning learnable latent spaces to arbitrary target distributions by leveraging flow-based generative models as priors. Our method first pretrains a flow model on the target features to capture the underlying distribution. This fixed flow model subsequently regularizes the latent space via an alignment loss, which reformulates the flow matching objective to treat the latents as optimization targets. We formally prove that minimizing this alignment loss establishes a computationally tractable surrogate objective for maximizing a variational lower bound on the log-likelihood of latents under the target distribution. Notably, the proposed method eliminates computationally expensive likelihood evaluations and avoids ODE solving during optimization. As a proof of concept, we demonstrate in a controlled setting that the alignment loss landscape closely approximates the negative log-likelihood of the target distribution. We further validate the effectiveness of our approach through large-scale image generation experiments on ImageNet with diverse target distributions, accompanied by detailed discussions and ablation studies. With both theoretical and empirical validation, our framework paves a new way for latent space alignment. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 16:59:53 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Li",
"Yizhuo",
""
],
[
"Ge",
"Yuying",
""
],
[
"Ge",
"Yixiao",
""
],
[
"Shan",
"Ying",
""
],
[
"Luo",
"Ping",
""
]
] |
2506.05241 | Qing Lyu | Qing Lyu and Mai Vu | Joint Beamforming and Integer User Association using a GNN with Gumbel-Softmax Reparameterizations | null | null | null | null | eess.SP | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Machine learning (ML) models can effectively optimize a multi-cell wireless network by designing the beamforming vectors and association decisions. Existing ML designs, however, often needs to approximate the integer association variables with a probability distribution output. We propose a novel graph neural network (GNN) structure that jointly optimize beamforming vectors and user association while guaranteeing association output as integers. The integer association constraints are satisfied using the Gumbel-Softmax (GS) reparameterization, without increasing computational complexity. Simulation results demonstrate that our proposed GS-based GNN consistently achieves integer association decisions and yields a higher sum-rate, especially when generalized to larger networks, compared to all other fractional association methods. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 17:01:27 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Lyu",
"Qing",
""
],
[
"Vu",
"Mai",
""
]
] |
2506.05242 | Haifeng Sun | Zhiqiang Wang, Haohua Du, Junyang Wang, Haifeng Sun, Kaiwen Guo, Haikuo Yu, Chao Liu, Xiang-Yang Li | SECNEURON: Reliable and Flexible Abuse Control in Local LLMs via Hybrid Neuron Encryption | null | null | null | null | cs.CR | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Large language models (LLMs) with diverse capabilities are increasingly being deployed in local environments, presenting significant security and controllability challenges. These locally deployed LLMs operate outside the direct control of developers, rendering them more susceptible to abuse. Existing mitigation techniques mainly designed for cloud-based LLM services are frequently circumvented or ineffective in deployer-controlled environments. We propose SECNEURON, the first framework that seamlessly embeds classic access control within the intrinsic capabilities of LLMs, achieving reliable, cost-effective, flexible, and certified abuse control for local deployed LLMs. SECNEURON employs neuron-level encryption and selective decryption to dynamically control the task-specific capabilities of LLMs, limiting unauthorized task abuse without compromising others. We first design a task-specific neuron extraction mechanism to decouple logically related neurons and construct a layered policy tree for handling coupled neurons. We then introduce a flexible and efficient hybrid encryption framework for millions of neurons in LLMs. Finally, we developed a distribution-based decrypted neuron detection mechanism on ciphertext to ensure the effectiveness of partially decrypted LLMs. We proved that SECNEURON satisfies IND-CPA Security and Collusion Resistance Security under the Task Controllability Principle. Experiments on various task settings show that SECNEURON limits unauthorized task accuracy to below 25% while keeping authorized accuracy loss with 2%. Using an unauthorized Code task example, the accuracy of abuse-related malicious code generation was reduced from 59% to 15%. SECNEURON also mitigates unauthorized data leakage, reducing PII extraction rates to below 5% and membership inference to random guesses. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 17:01:28 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Wang",
"Zhiqiang",
""
],
[
"Du",
"Haohua",
""
],
[
"Wang",
"Junyang",
""
],
[
"Sun",
"Haifeng",
""
],
[
"Guo",
"Kaiwen",
""
],
[
"Yu",
"Haikuo",
""
],
[
"Liu",
"Chao",
""
],
[
"Li",
"Xiang-Yang",
""
]
] |
2506.05243 | Ron Eliav | Ron Eliav, Arie Cattan, Eran Hirsch, Shahaf Bassan, Elias Stengel-Eskin, Mohit Bansal, Ido Dagan | CLATTER: Comprehensive Entailment Reasoning for Hallucination Detection | null | null | null | null | cs.CL | http://creativecommons.org/licenses/by/4.0/ | A common approach to hallucination detection casts it as a natural language inference (NLI) task, often using LLMs to classify whether the generated text is entailed by corresponding reference texts. Since entailment classification is a complex reasoning task, one would expect that LLMs could benefit from generating an explicit reasoning process, as in CoT reasoning or the explicit ``thinking'' of recent reasoning models. In this work, we propose that guiding such models to perform a systematic and comprehensive reasoning process -- one that both decomposes the text into smaller facts and also finds evidence in the source for each fact -- allows models to execute much finer-grained and accurate entailment decisions, leading to increased performance. To that end, we define a 3-step reasoning process, consisting of (i) claim decomposition, (ii) sub-claim attribution and entailment classification, and (iii) aggregated classification, showing that such guided reasoning indeed yields improved hallucination detection. Following this reasoning framework, we introduce an analysis scheme, consisting of several metrics that measure the quality of the intermediate reasoning steps, which provided additional empirical evidence for the improved quality of our guided reasoning scheme. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 17:02:52 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Eliav",
"Ron",
""
],
[
"Cattan",
"Arie",
""
],
[
"Hirsch",
"Eran",
""
],
[
"Bassan",
"Shahaf",
""
],
[
"Stengel-Eskin",
"Elias",
""
],
[
"Bansal",
"Mohit",
""
],
[
"Dagan",
"Ido",
""
]
] |
2506.05244 | Hao Zhang | Hao Zhang, Alex Kamenev | How to Train Your Dragon: Quantum Neural Networks | 7 pages, 5 figures | null | null | null | quant-ph cond-mat.dis-nn | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Training of neural networks (NNs) has emerged as a major consumer of both computational and energy resources. We demonstrate that quantum annealing platforms, such as D-Wave, can enable fast and efficient training of classical NNs, which are then deployable on conventional hardware. From a physics perspective, NN training can be viewed as a dynamical phase transition: the system evolves from an initial spin glass state to a highly ordered, trained state. This process involves eliminating numerous undesired minima in its energy landscape--akin to cutting off the ever-regenerating heads of a dragon. The advantage of annealing devices is their ability to rapidly find multiple deep states (dragon heads to be cut). We found that this quantum-assisted training achieves superior performance scaling compared to classical backpropagation methods, with a notably higher scaling exponent (1.01 vs. 0.78). It may be further increased up to a factor of 2 with a fully coherent quantum platform using a variant of the Grover algorithm. Furthermore, we argue that even a modestly sized annealer can be beneficial to train a deep NN by being applied sequentially to a few layers at a time. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 17:03:26 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Zhang",
"Hao",
""
],
[
"Kamenev",
"Alex",
""
]
] |
2506.05245 | Shaoxuan Chen | Shaoxuan Chen, Su Yang, Panayotis G. Kevrekidis, Wei Zhu | Robust Moment Identification for Nonlinear PDEs via a Neural ODE Approach | null | null | null | null | nlin.PS cs.LG | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | We propose a data-driven framework for learning reduced-order moment dynamics from PDE-governed systems using Neural ODEs. In contrast to derivative-based methods like SINDy, which necessitate densely sampled data and are sensitive to noise, our approach based on Neural ODEs directly models moment trajectories, enabling robust learning from sparse and potentially irregular time series. Using as an application platform the nonlinear Schr\"{o}dinger equation, the framework accurately recovers governing moment dynamics when closure is available, even with limited and irregular observations. For systems without analytical closure, we introduce a data-driven coordinate transformation strategy based on Stiefel manifold optimization, enabling the discovery of low-dimensional representations in which the moment dynamics become closed, facilitating interpretable and reliable modeling. We also explore cases where a closure model is not known, such as a Fisher-KPP reaction-diffusion system. Here we demonstrate that Neural ODEs can still effectively approximate the unclosed moment dynamics and achieve superior extrapolation accuracy compared to physical-expert-derived ODE models. This advantage remains robust even under sparse and irregular sampling, highlighting the method's robustness in data-limited settings. Our results highlight the Neural ODE framework as a powerful and flexible tool for learning interpretable, low-dimensional moment dynamics in complex PDE-governed systems. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 17:03:42 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Chen",
"Shaoxuan",
""
],
[
"Yang",
"Su",
""
],
[
"Kevrekidis",
"Panayotis G.",
""
],
[
"Zhu",
"Wei",
""
]
] |
2506.05246 | Daniel Remenik | Jonas Arista, Daniel Remenik, Avelio Sep\'ulveda | Myopic non-intersection in a periodic potential | 26 pages, 3 figures | null | null | null | math.PR math-ph math.MP | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | We introduce a class of Markov processes conditioned to avoid intersection over a moving time window of length T>0, a setting we refer to as myopic non-intersection. In particular, we study a system of myopic non-intersecting Brownian motions subject to a periodic potential. Our focus lies in understanding the interplay between the confining effect of the potential and the repulsion induced by the non-intersection constraint. We show that, in the long time limit, and as both T and the strength of the potential become large, the model converges to a system of myopic non-intersecting random walks, which transitions between standard non-intersection dynamics and exclusion behavior. The main technical contribution of the paper is the introduction of an algorithm, based on a modification of the acceptance-rejection sampling scheme, that provides an explicit construction of myopically constrained systems. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 17:05:27 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Arista",
"Jonas",
""
],
[
"Remenik",
"Daniel",
""
],
[
"Sepúlveda",
"Avelio",
""
]
] |
2506.05247 | Takato Yoshimura | Takato Yoshimura and \v{Z}iga Krajnik | Hydrodynamic fluctuations of stochastic charged cellular automata | 7+6 pages, 2 figures | null | null | null | cond-mat.stat-mech cond-mat.str-el hep-th math-ph math.MP | http://creativecommons.org/licenses/by/4.0/ | We study charge fluctuations of a family of stochastic charged cellular automata away from the deterministic single-file limit and obtain the exact typical charge probability distributions, known to be anomalous, using hydrodynamics. The cellular automata considered are examples of linearly degenerate systems where two distinct mechanisms of diffusion, namely normal and convective diffusion, coexist. Our formalism, based on macroscopic fluctuation theory, allows us to describe current fluctuations stemming from these two diffusive processes, and we expect it to be applicable to generic linearly degenerate systems. The derived probability distributions match the exact microscopic result and numerical simulations. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 17:07:31 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Yoshimura",
"Takato",
""
],
[
"Krajnik",
"Žiga",
""
]
] |
2506.05248 | Jonathan Monta\~no | Steven Dale Cutkosky and Jonathan Monta\~no | Degree functions of graded families of ideals | null | null | null | null | math.AC | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | We express multiplicities and degree functions of graded families of $\mathfrak{m}_R$-primary ideals in an excellent normal local ring $(R,\mathfrak{m}_R)$ as limits of intersection products. Moreover, in dimension 2, we show more refined results for divisorial filtrations. Finally, also in dimension 2, we give an example of a non-Noetherian divisorial filtration $\{I_n\}_{n\geqslant 0}$ of $\mathfrak{m}_R$-primary ideals such that the union of all the sets of Rees valuations of all the $I_n$ is a finite set, and another example of a (necessarily non-Noetherian) divisorial filtration of $\mathfrak{m}_R$-primary ideals such that the set of all Rees valuations is infinite. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 17:09:06 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Cutkosky",
"Steven Dale",
""
],
[
"Montaño",
"Jonathan",
""
]
] |
2506.05249 | Zhen Qin | Zhen Qin and Jinxin Zhou and Zhihui Zhu | On the Convergence of Gradient Descent on Learning Transformers with Residual Connections | null | null | null | null | cs.LG math.OC | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Transformer models have emerged as fundamental tools across various scientific and engineering disciplines, owing to their outstanding performance in diverse applications. Despite this empirical success, the theoretical foundations of Transformers remain relatively underdeveloped, particularly in understanding their training dynamics. Existing research predominantly examines isolated components--such as self-attention mechanisms and feedforward networks--without thoroughly investigating the interdependencies between these components, especially when residual connections are present. In this paper, we aim to bridge this gap by analyzing the convergence behavior of a structurally complete yet single-layer Transformer, comprising self-attention, a feedforward network, and residual connections. We demonstrate that, under appropriate initialization, gradient descent exhibits a linear convergence rate, where the convergence speed is determined by the minimum and maximum singular values of the output matrix from the attention layer. Moreover, our analysis reveals that residual connections serve to ameliorate the ill-conditioning of this output matrix, an issue stemming from the low-rank structure imposed by the softmax operation, thereby promoting enhanced optimization stability. We also extend our theoretical findings to a multi-layer Transformer architecture, confirming the linear convergence rate of gradient descent under suitable initialization. Empirical results corroborate our theoretical insights, illustrating the beneficial role of residual connections in promoting convergence stability. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 17:10:22 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Qin",
"Zhen",
""
],
[
"Zhou",
"Jinxin",
""
],
[
"Zhu",
"Zhihui",
""
]
] |
2506.05250 | Zhiyun Deng | Zhiyun Deng, Dongmyeong Lee, Amanda Adkins, Jesse Quattrociocchi, Christian Ellis, Joydeep Biswas | Spatiotemporal Contrastive Learning for Cross-View Video Localization in Unstructured Off-road Terrains | null | null | null | null | cs.CV cs.RO | http://creativecommons.org/licenses/by/4.0/ | Robust cross-view 3-DoF localization in GPS-denied, off-road environments remains challenging due to (1) perceptual ambiguities from repetitive vegetation and unstructured terrain, and (2) seasonal shifts that significantly alter scene appearance, hindering alignment with outdated satellite imagery. To address this, we introduce MoViX, a self-supervised cross-view video localization framework that learns viewpoint- and season-invariant representations while preserving directional awareness essential for accurate localization. MoViX employs a pose-dependent positive sampling strategy to enhance directional discrimination and temporally aligned hard negative mining to discourage shortcut learning from seasonal cues. A motion-informed frame sampler selects spatially diverse frames, and a lightweight temporal aggregator emphasizes geometrically aligned observations while downweighting ambiguous ones. At inference, MoViX runs within a Monte Carlo Localization framework, using a learned cross-view matching module in place of handcrafted models. Entropy-guided temperature scaling enables robust multi-hypothesis tracking and confident convergence under visual ambiguity. We evaluate MoViX on the TartanDrive 2.0 dataset, training on under 30 minutes of data and testing over 12.29 km. Despite outdated satellite imagery, MoViX localizes within 25 meters of ground truth 93% of the time, and within 50 meters 100% of the time in unseen regions, outperforming state-of-the-art baselines without environment-specific tuning. We further demonstrate generalization on a real-world off-road dataset from a geographically distinct site with a different robot platform. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 17:10:29 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Deng",
"Zhiyun",
""
],
[
"Lee",
"Dongmyeong",
""
],
[
"Adkins",
"Amanda",
""
],
[
"Quattrociocchi",
"Jesse",
""
],
[
"Ellis",
"Christian",
""
],
[
"Biswas",
"Joydeep",
""
]
] |
2506.05251 | J. Carlos Martinez Mori | J. Carlos Mart\'inez Mori and Alejandro Toriello | Cooperation and the Design of Public Goods | 26th ACM Conference on Economics and Computation (EC '25) | null | null | null | cs.GT math.OC | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | We consider the cooperative elements that arise in the design of public goods, such as transportation policies and infrastructure. These involve a variety of stakeholders: governments, businesses, advocates, and users. Their eventual deployment depends on the decision maker's ability to garner sufficient support from each of these groups; we formalize these strategic requirements from the perspective of cooperative game theory. Specifically, we introduce non-transferable utility, linear production (NTU LP) games, which combine the game-theoretic tensions inherent in public decision-making with the modeling flexibility of linear programming. We derive structural properties regarding the non-emptiness, representability and complexity of the core, a solution concept that models the viability of cooperation. In particular, we provide fairly general sufficient conditions under which the core of an NTU LP game is guaranteed to be non-empty, prove that determining membership in the core is co-NP-complete, and develop a cutting plane algorithm to optimize various social welfare objectives subject to core membership. Lastly, we apply these results in a data-driven case study on service plan optimization for the Chicago bus system. As our study illustrates, cooperation is necessary for the successful deployment of transportation service plans and similar public goods, but it may also have adverse or counterintuitive distributive implications. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 17:11:53 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Mori",
"J. Carlos Martínez",
""
],
[
"Toriello",
"Alejandro",
""
]
] |
2506.05252 | Alec Sun | Dravyansh Sharma, Alec Sun | Conservative classifiers do consistently well with improving agents: characterizing statistical and online learning | 24 pages | null | null | null | cs.LG cs.GT cs.MA | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | Machine learning is now ubiquitous in societal decision-making, for example in evaluating job candidates or loan applications, and it is increasingly important to take into account how classified agents will react to the learning algorithms. The majority of recent literature on strategic classification has focused on reducing and countering deceptive behaviors by the classified agents, but recent work of Attias et al. identifies surprising properties of learnability when the agents genuinely improve in order to attain the desirable classification, such as smaller generalization error than standard PAC-learning. In this paper we characterize so-called learnability with improvements across multiple new axes. We introduce an asymmetric variant of minimally consistent concept classes and use it to provide an exact characterization of proper learning with improvements in the realizable setting. While prior work studies learnability only under general, arbitrary agent improvement regions, we give positive results for more natural Euclidean ball improvement sets. In particular, we characterize improper learning under a mild generative assumption on the data distribution. We further show how to learn in more challenging settings, achieving lower generalization error under well-studied bounded noise models and obtaining mistake bounds in realizable and agnostic online learning. We resolve open questions posed by Attias et al. for both proper and improper learning. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 17:13:59 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Sharma",
"Dravyansh",
""
],
[
"Sun",
"Alec",
""
]
] |
2506.05253 | Christian Jepsen | Christian Baadsgaard Jepsen | The Veneziano Amplitude in any Dimension and a Virasoro-Shapiro Partial Amplitude | 17 pages, 2 figures | null | null | null | hep-th | http://creativecommons.org/licenses/by/4.0/ | This paper demonstrates how the Veneziano partial amplitude of bosonic string theory admits a generalization to world-(hyper)surfaces of any dimension $d$. In particular, for $d=2$, by carving up the worldsheet integral according to stipulations imposed on conformal cross ratios, the Virasoro-Shapiro full amplitude can be decomposed into a sum of three partial amplitudes. The amplitudes obtained on generalizing the Veneziano amplitude all contain tachyons. To explore candidate tachyon-free and supersymmetric versions of these amplitudes, a new bootstrap principle is introduced and applied, which demands that towers of residues alternate between all-even and all-odd spin partial waves. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 17:14:28 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Jepsen",
"Christian Baadsgaard",
""
]
] |
2506.05254 | Vefa Goksel | Robert L. Benedetto and Vefa Goksel | The non-unit conjecture for Misiurewicz parameters | 19 pages | null | null | null | math.NT | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | A Misiurewicz parameter is a complex number $c$ for which the orbit of the critical point $z=0$ under $z^2+c$ is strictly preperiodic. Such parameters play the same role in dynamical moduli spaces as singular moduli (corresponding to CM elliptic curves) play on modular curves. Building on our earlier work, we investigate whether the difference of two Misiurewicz parameters can be an algebraic unit. (The corresponding question for singular moduli was recently answered in the negative by Li.) We answer this dynamical question in many new cases under a widely believed irreducibility assumption. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 17:15:46 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Benedetto",
"Robert L.",
""
],
[
"Goksel",
"Vefa",
""
]
] |
2506.05255 | Paolo Facchi | Giuliano Angelone, Elisa Ercolessi, Paolo Facchi, Rocco Maggi, Giuseppe Marmo, Saverio Pascazio, Francesco V. Pepe | Electromagnetism: an intrinsic approach to Hadamard's method of descent | 35 pages | null | null | null | math-ph math.MP physics.class-ph | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | We present a systematic geometric framework for the dimensional reduction of classical electromagnetism based on the concept of descent along vector fields of invariance. By exploring the interplay between the Lie derivative and the Hodge star operator, we implement descent conditions on differential forms that reduce Maxwell's equations in four-dimensional spacetime to electromagnetic theories in lower dimensions. We also consider multiple descent along pairwise commuting vector fields of invariance, yielding a finer decomposition of Maxwell's equations. Our results provide a unified and geometrically transparent interpretation of dimensional reduction, with potential applications to field theories in lower-dimensional spacetimes. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 17:17:01 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Angelone",
"Giuliano",
""
],
[
"Ercolessi",
"Elisa",
""
],
[
"Facchi",
"Paolo",
""
],
[
"Maggi",
"Rocco",
""
],
[
"Marmo",
"Giuseppe",
""
],
[
"Pascazio",
"Saverio",
""
],
[
"Pepe",
"Francesco V.",
""
]
] |
2506.05256 | Violet Xiang | Violet Xiang, Chase Blagden, Rafael Rafailov, Nathan Lile, Sang Truong, Chelsea Finn, Nick Haber | Just Enough Thinking: Efficient Reasoning with Adaptive Length Penalties Reinforcement Learning | null | null | null | null | cs.AI cs.LG | http://creativecommons.org/licenses/by/4.0/ | Large reasoning models (LRMs) achieve higher performance on challenging reasoning tasks by generating more tokens at inference time, but this verbosity often wastes computation on easy problems. Existing solutions, including supervised finetuning on shorter traces, user-controlled budgets, or RL with uniform penalties, either require data curation, manual configuration, or treat all problems alike regardless of difficulty. We introduce Adaptive Length Penalty (ALP), a reinforcement learning objective tailoring generation length to per-prompt solve rate. During training, ALP monitors each prompt's online solve rate through multiple rollouts and adds a differentiable penalty whose magnitude scales inversely with that rate, so confident (easy) prompts incur a high cost for extra tokens while hard prompts remain unhindered. Posttraining DeepScaleR-1.5B with ALP cuts average token usage by 50\% without significantly dropping performance. Relative to fixed-budget and uniform penalty baselines, ALP redistributes its reduced budget more intelligently by cutting compute on easy prompts and reallocating saved tokens to difficult ones, delivering higher accuracy on the hardest problems with higher cost. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 17:17:05 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Xiang",
"Violet",
""
],
[
"Blagden",
"Chase",
""
],
[
"Rafailov",
"Rafael",
""
],
[
"Lile",
"Nathan",
""
],
[
"Truong",
"Sang",
""
],
[
"Finn",
"Chelsea",
""
],
[
"Haber",
"Nick",
""
]
] |
2506.05257 | Alfie Davies | Alfie Davies, Sarah Miller, Rebecca Milley | On sums of $\mathscr{P}$-free forms under mis\`ere play | 35 pages, 2 figures | null | null | null | math.CO | http://creativecommons.org/licenses/by/4.0/ | Milley and Renault proved an interesting characterisation of invertible elements in the dead-ending universe: they are the games with no subpositions of outcome $\mathscr{P}$ (the '$\mathscr{P}$-free' games). We generalise their approach to obtain a stronger result and show in particular that the set of $\mathscr{P}$-free blocking games is closed under addition, which yields that every $\mathscr{P}$-free blocking game is invertible modulo the blocking universe. This has consequences for the invertible subgroups of various other mis\`ere monoids. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 17:19:54 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Davies",
"Alfie",
""
],
[
"Miller",
"Sarah",
""
],
[
"Milley",
"Rebecca",
""
]
] |
2506.05258 | David Tucker-Smith | Snehadri Das, Will Howe, Brian Shuve, David Tucker-Smith, and Ruby Yager | Closing in on singly charged scalars | 13 pages, 6 figures | null | null | null | hep-ph hep-ex | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ | We investigate current experimental constraints and future search prospects for a hypothetical spin-zero particle that carries unit electric charge: a singly charged scalar (SCS). In addition to providing useful benchmarks for collider searches, SCS particles are also well-motivated ingredients in relatively minimal dark sectors. We focus on scenarios in which the SCS decays promptly at colliders to a lepton plus either a neutrino or an invisible dark-sector particle of negligible mass. A promptly decaying SCS can easily have appreciable branching ratios to more than one lepton flavor while remaining consistent with constraints on lepton flavor violation. This broadens the allowed range of SCS masses to extend well beyond those for right-handed selectrons, smuons, or staus. For particular benchmark SCS branching ratios, we find that SCS masses above ~185 GeV and in a lower-mass window ~80-125 GeV are still allowed at 95% confidence level. We carry out Monte Carlo simulations to explore the potential of a boosted-decision-tree-based analysis to probe the surviving SCS parameter space in future searches at the (HL-)LHC, finding a significant increase in sensitivity relative to cut-based analyses both in the lower-mass window and at higher SCS masses. | [
{
"version": "v1",
"created": "Thu, 05 Jun 2025 17:20:03 GMT"
}
] | 2025-06-06T00:00:00 | [
[
"Das",
"Snehadri",
""
],
[
"Howe",
"Will",
""
],
[
"Shuve",
"Brian",
""
],
[
"Tucker-Smith",
"David",
""
],
[
"Yager",
"Ruby",
""
]
] |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.