resnet18-cifar10-unlearning

Model Card for jaeunglee/resnet18-cifar10-unlearning

This repository contains ResNet18 models retrained on the CIFAR-10 dataset with specific classes excluded during training. Each model is trained to study the impact of class exclusion on model performance and generalization.

Paper: Unlearning Comparator: A Visual Analytics System for Comparative Evaluation of Machine Unlearning Methods

Project Page: https://gnueaj.github.io/Machine-Unlearning-Comparator/

GitHub Repository: https://github.com/gnueaj/Machine-Unlearning-Comparator

---

Evaluation

• Testing Data: CIFAR-10 test set
• Metrics: Top-1 accuracy

Results

Training Details

Training Procedure

• Base Model: ResNet18
• Dataset: CIFAR-10
• Excluded Class: Varies by model
• Loss Function: CrossEntropyLoss
• Optimizer: SGD with:
  • Learning rate: 0.1
  • Momentum: 0.9
  • Weight decay: 5e-4
  • Nesterov: True
• Scheduler: CosineAnnealingLR (T_max: 200)
• Training Epochs: 200
• Batch Size: 128
• Hardware: Single GPU

Notes on Training

The training recipe is adapted from the paper "Benchopt: Reproducible, efficient and collaborative optimization benchmarks", which provides a reproducible and optimized setup for training ResNet18 on the CIFAR-10 dataset. This ensures that the training methodology aligns with established benchmarks for reproducibility and comparability.

Data Preprocessing

The following transformations were applied to the CIFAR-10 dataset:

• Base Transformations (applied to both training and test sets):

  • Conversion to PyTorch tensors using ToTensor().
  • Normalization using mean (0.4914, 0.4822, 0.4465) and standard deviation (0.2023, 0.1994, 0.2010).

• Training Set Augmentation (only for training data):

  • RandomCrop(32, padding=4): Randomly crops images with padding for spatial variation.
  • RandomHorizontalFlip(): Randomly flips images horizontally with a 50% probability.

These augmentations help improve the model's ability to generalize by introducing variability in the training data.

Model Description

• Developed by: Jaeung Lee
• Model type: Image Classification
• License: MIT

Related Work

This model is part of the research conducted using the Machine Unlearning Comparator. The tool was developed to compare various machine unlearning methods and their effects on models.

Uses

Direct Use

These models can be directly used for evaluating the effect of excluding specific classes from the CIFAR-10 dataset during training.

Out-of-Scope Use

The models are not suitable for tasks requiring general-purpose image classification beyond the CIFAR-10 dataset.

How to Get Started with the Model

Use the code below to load the models with the appropriate architecture and weights:

import torch
import torch.nn as nn
from torchvision import models

def get_resnet18(num_classes=10):
    model = models.resnet18(weights=None)
    model.conv1 = nn.Conv2d(3, 64, kernel_size=3, stride=1, padding=1, bias=False)
    model.maxpool = nn.Identity()
    model.fc = nn.Linear(model.fc.in_features, num_classes)
    return model

# Load a pretrained model
def load_model(model_path, num_classes=10):
    model = get_resnet18(num_classes=num_classes)
    model.load_state_dict(torch.load(model_path))
    return model

# Example usage
model = load_model("resnet18_cifar10_no_airplane.pth", num_classes=10)

Citation

If you use this repository or its models in your work, please consider citing it:

📄 Paper

Unlearning Comparator: A Visual Analytics System for Comparative Evaluation of Machine Unlearning Methods

APA: Jaeung Lee. (2024). ResNet18 Models Trained on CIFAR-10 with Class Exclusion. Retrieved from https://huggingface.co/jaeunglee/resnet18-cifar10-unlearn

License

This repository is shared under the MIT License.