mnist-interpretable-ml

This model is trained to classify handwritten digits from the MNIST dataset while attempting to explain its predictions based on learned features. It is built upon the concepts introduced in the Self-Explaining Neural Networks (SENN) paper.

Model Details

• Task: Image Classification & Interpretable Machine Learning
• Dataset: MNIST (split into 50,000 training samples and 10,000 validation samples).
• Format: PyTorch Checkpoint (.pth)

How to Use

Since the weights on the repository are saved as senn_full_mnist.pth[cite: 4] (a raw PyTorch state dictionary checkpoint), you must have the original network architecture code from the project's GitHub repository to load it properly.

Acknowledgments & References

This project is built upon the following materials and source code:

Methodology & Paper

• Towards Robust Interpretability with Self-Explaining Neural Networks - David Alvarez-Melis, Tommi S. Jaakkola. arXiv:1806.07538

Source Code

• Project Repository: duckling2211/mnist-interpretable-ml