nneurotransmitterr / aai-project-mnist-stable-classifier

CSE5001 Artificial Intelligence 2023 Fall Final Project: Train a classifier for modified MNIST with unstable (spurious) features.

AAI Fall 2023 Project: Train a Stable MNIST Classifier

Train a classifier for modified MNIST with unstable (spurious) features.

How to use

1. Install the requirements in requirements.txt.
2. Extract the data to the processed_data/ folder.
3. Run the test script with python test.py.

Optional

• Inspect data with data_inspection.ipynb.
• Run the training script with python main.py.

Directory Structure

aai-project-mnist-stable-classifier/
├── processed_data/
│   ├── train/
│   ├── test/
│   └── val/
├── models/
│   ├── CNN.py
│   ├── MLP.py
│   └── OCNN.py
├── saved_models/
│   ├── mlp_10000_20240106.pth
│   └── OriginalMNISTClassifier.pth
├── data_inspection.ipynb
├── OddMNIST.py
├── main.py
├── test.py
├── results.txt
├── requirements.txt
├── LICENSE
└── README.md

References

Codes for these papers are used in this project:

1. InvariantRiskMinimization

@article{InvariantRiskMinimization,
    title={Invariant Risk Minimization},
    author={Arjovsky, Martin and Bottou, L{\'e}on and Gulrajani, Ishaan and Lopez-Paz, David},
    journal={arXiv},
    year={2019}
}