Update README.md
Browse files
README.md
CHANGED
|
@@ -38,6 +38,16 @@ LWM-v1.1 is designed to be seamlessly integrated into downstream tasks as a sour
|
|
| 38 |
- **Broad Generalization**: Trained on a larger, more diverse dataset for reliable performance across environments.
|
| 39 |
- **Task Adaptability**: Fine-tuning options enable seamless integration into a wide range of applications.
|
| 40 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 41 |
---
|
| 42 |
|
| 43 |
## **Overview of Main Changes in LWM-v1.1**
|
|
@@ -475,7 +485,22 @@ chs = lwm_inference(
|
|
| 475 |
)
|
| 476 |
```
|
| 477 |
|
| 478 |
-
This generates embeddings or visualizations, depending on your configuration.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 479 |
|
| 480 |
---
|
| 481 |
|
|
|
|
| 38 |
- **Broad Generalization**: Trained on a larger, more diverse dataset for reliable performance across environments.
|
| 39 |
- **Task Adaptability**: Fine-tuning options enable seamless integration into a wide range of applications.
|
| 40 |
|
| 41 |
+
For example, the following figure demonstrates the advantages of using **LWM-v1.1-based highly compact CLS embeddings** and **high-dimensional channel embeddings** over raw channels for the LoS/NLoS classification task. The raw dataset is derived from channels of size (128, 32) between BS 3 and 8,299 users in the densified Denver scenario of the DeepMIMO dataset.
|
| 42 |
+
|
| 43 |
+
<p align="center">
|
| 44 |
+
<img src="https://huggingface.co/wi-lab/lwm-v1.1/resolve/main/images/los_perf.png" alt="LoS/NLoS Classification Performance" width="800"/>
|
| 45 |
+
</p>
|
| 46 |
+
|
| 47 |
+
<p align="center">
|
| 48 |
+
<strong>Figure:</strong> LoS/NLoS classification performance comparison
|
| 49 |
+
</p>
|
| 50 |
+
|
| 51 |
---
|
| 52 |
|
| 53 |
## **Overview of Main Changes in LWM-v1.1**
|
|
|
|
| 485 |
)
|
| 486 |
```
|
| 487 |
|
| 488 |
+
This generates embeddings or visualizations, depending on your configuration. For example, the following figures show the 2D T-SNE representations of original, embedding, and fine-tuned embedding spaces for the LoS/NLoS classification and beam prediction tasks.
|
| 489 |
+
|
| 490 |
+
### **LoS/NLoS Classification Task**
|
| 491 |
+
|
| 492 |
+
|  |  |  |
|
| 493 |
+
|:---------------------------------------------:|:---------------------------------------------:|:---------------------------------------------:|
|
| 494 |
+
| **Raw Channels** | **General-purpose Embeddings** | **Task-specific Embeddings** |
|
| 495 |
+
|
| 496 |
+
---
|
| 497 |
+
|
| 498 |
+
### **Beam Prediction Task**
|
| 499 |
+
|
| 500 |
+
|  |  |  |
|
| 501 |
+
|:---------------------------------------------:|:---------------------------------------------:|:---------------------------------------------:|
|
| 502 |
+
| **Raw Channels** | **General-purpose Embeddings** | **Task-specific Embeddings** |
|
| 503 |
+
|
| 504 |
|
| 505 |
---
|
| 506 |
|