Yurim0507/SALUN

Evaluation Report

Testing Data

Dataset: CIFAR-10 Test Set
Metrics: Top-1 Accuracy

---

Training Details

Training Procedure

• Base Model: ResNet18
• Dataset: CIFAR-10
• Excluded Class: Varies by model
• Loss Function: Negative Log-Likelihood Loss
• Optimizer: SGD with:
  • Learning rate: 0.01
  • Momentum: 0.9
  • Weight decay: 5e-4
  • Nesterov: True
• Training Epochs: 8
• Batch Size: 512
• Hardware: Single GPU (NVIDIA GeForce RTX 3090)

SALUN Specifics

• Threshold: 0.1

Algorithm

The SALUN (Saliency Unlearning) algorithm was used for inexact unlearning. This method involves impairing the influence of a specific class and fine-tuning the model to regain its accuracy for the remaining classes.

Each resulting model (cifar10_resnet18_SALUN_X.pth) corresponds to a scenario where a single class (X) has been unlearned. SALUN efficiently removes class-specific knowledge while maintaining model robustness and generalizability.

For more details on the SALUN algorithm, refer to the GitHub repository.

---

Results

Model	Excluded Class	Forget class acc(loss)	Retain class acc(loss)
cifar10_resnet18_SALUN_0.pth	Airplane	0.7 (2.550)	90.00 (0.347)
cifar10_resnet18_SALUN_1.pth	Automobile	0.0 (2.976)	88.73 (0.404)
cifar10_resnet18_SALUN_2.pth	Bird	2.4 (2.862)	89.13 (0.356)
cifar10_resnet18_SALUN_3.pth	Cat	0.0 (3.640)	92.13 (0.262)
cifar10_resnet18_SALUN_4.pth	Deer	0.9 (2.749)	89.74 (0.349)
cifar10_resnet18_SALUN_5.pth	Dog	3.7 (2.870)	88.92 (0.363)
cifar10_resnet18_SALUN_6.pth	Frog	1.7 (3.236)	86.23 (0.486)
cifar10_resnet18_SALUN_7.pth	Horse	0.0 (3.119)	90.03 (0.342)
cifar10_resnet18_SALUN_8.pth	Ship	9.4 (2.685)	90.86 (0.320)
cifar10_resnet18_SALUN_9.pth	Truck	0.5 (2.748)	89.88 (0.362)