Update README.md

README.md

@@ -52,27 +52,51 @@ For more details on the SALUN algorithm, refer to the [GitHub repository](https:
 
 ## Results
 
-| Model                              | Excluded Class | CIFAR-10 Accuracy (%) |
-|------------------------------------|----------------|-----------------------|
-| cifar10_resnet18_SALUN_0.pth       | Airplane       | 72.31                |
-| cifar10_resnet18_SALUN_1.pth       | Automobile     | 69.45                |
-| cifar10_resnet18_SALUN_2.pth       | Bird           | 67.35                |
-| cifar10_resnet18_SALUN_3.pth       | Cat            | 77.85                |
-| cifar10_resnet18_SALUN_4.pth       | Deer           | 65.14                |
-| cifar10_resnet18_SALUN_5.pth       | Dog            | 74.00                |
-| cifar10_resnet18_SALUN_6.pth       | Frog           | 71.00                |
-| cifar10_resnet18_SALUN_7.pth       | Horse          | 73.44                |
-| cifar10_resnet18_SALUN_8.pth       | Ship           | 70.13                |
-| cifar10_resnet18_SALUN_9.pth       | Truck          | 73.10                |
+| Model                              | Excluded Class | Forget class acc(loss) | Retain class acc(loss) |
+|------------------------------------|----------------|-------------------------|-------------------------|
+| cifar10_resnet18_SALUN_0.pth       | Airplane       | 0.1 (3.274)            | 80.33 (0.589)          |
+| cifar10_resnet18_SALUN_1.pth       | Automobile     | 0.0 (3.373)            | 77.17 (0.731)          |
+| cifar10_resnet18_SALUN_2.pth       | Bird           | 0.3 (3.407)            | 74.80 (0.801)          |
+| cifar10_resnet18_SALUN_3.pth       | Cat            | 0.0 (3.688)            | 86.50 (0.423)          |
+| cifar10_resnet18_SALUN_4.pth       | Deer           | 0.0 (2.918)            | 72.38 (0.839)          |
+| cifar10_resnet18_SALUN_5.pth       | Dog            | 0.0 (3.713)            | 82.22 (0.553)          |
+| cifar10_resnet18_SALUN_6.pth       | Frog           | 0.0 (4.381)            | 78.43 (0.651)          |
+| cifar10_resnet18_SALUN_7.pth       | Horse          | 0.3 (3.056)            | 81.57 (0.560)          |
+| cifar10_resnet18_SALUN_8.pth       | Ship           | 0.6 (4.411)            | 77.86 (0.686)          |
+| cifar10_resnet18_SALUN_9.pth       | Truck          | 0.0 (4.993)            | 81.22 (0.578)          |
 
 ---
 
-## Notes
-- The **Top-1 Accuracy** metric represents the percentage of correctly classified samples from the CIFAR-10 test set.
-- The excluded class refers to the class omitted during model training to evaluate its effect on accuracy.
-- The average accuracy across all models is **71.77%**, with the highest accuracy observed for **Cat exclusion (77.85%)** and the lowest for **Deer exclusion (65.14%)**.
+### Notes
+
+1. **Forget Class Accuracy and Loss**:
+   - The forget class accuracy is consistently near zero across all excluded classes, indicating the effectiveness of the **Selective Synapse Dampening** method in excluding the target class. The highest forget class accuracy is `0.6%` for "Ship," while most classes report `0.0%`.
+   - The forget class loss varies, with "Truck" showing the highest loss (4.993) and "Deer" the lowest (2.918). This variability suggests differences in how strongly certain classes are suppressed.
+
+2. **Retain Class Accuracy and Loss**:
+   - Retain class accuracy is reasonably high across most classes, with "Cat" achieving the highest accuracy (86.50%) and "Deer" the lowest (72.38%). This shows that the model generally performs well on the retained classes.
+   - Retain class loss is minimal across all classes, with the lowest being "Cat" (0.423) and the highest being "Deer" (0.839). The low losses indicate stable performance on the retained classes.
+
+3. **Class-Specific Observations**:
+   - Some classes, such as "Ship" and "Horse," exhibit slightly higher forget class accuracies (0.6% and 0.3%, respectively), suggesting minor challenges in completely excluding these classes.
+   - The variability in retain class accuracy and forget class loss indicates that some classes are inherently harder to balance during the selective exclusion process.
 
 ---
 
-## Conclusion
-This report demonstrates the effectiveness of the SALUN algorithm for inexact unlearning on the CIFAR-10 dataset. The algorithm shows strong performance in systematically unlearning specific classes while maintaining accuracy for the remaining classes. Further validation with larger and more complex datasets (e.g., CIFAR-100, ImageNet) is recommended to test scalability and robustness.
+### Conclusion
+
+The results demonstrate the robustness of the **Selective Synapse Dampening** approach for excluding specific classes while maintaining high performance on the retained classes. However, certain trends and challenges emerge:
+
+- **Strengths**:
+  - The forget class accuracy is near zero for most classes, achieving effective exclusion of the target class.
+  - Retain class accuracy remains high (>70%) across all classes, with minimal loss, indicating strong performance on the retained tasks.
+
+- **Weaknesses**:
+  - Minor residual accuracy for excluded classes like "Ship" (0.6%) suggests room for improvement in class suppression.
+  - The retain class accuracy for "Deer" (72.38%) is noticeably lower compared to other classes, indicating potential trade-offs in performance for certain categories.
+
+- **Future Work**:
+  - Further refinement in suppression mechanisms could improve the forget class accuracy for challenging classes like "Ship" and "Horse."
+  - Adaptive techniques to balance performance across all retained classes may help address discrepancies in retain class accuracy (e.g., "Deer").
+  - A deeper analysis into class-specific characteristics could guide targeted optimization, making the model more robust for practical applications.
+