📚 [GitHub Repository] | 📝 [Paper]
Hello World, Welcome to the ELMAGIC PyTorch Experiments Repository!
This repository contains PyTorch code to reproduce the key experiments and results presented in the paper: ELMAGIC: Energy-Efficient Lean Model for Reliable Medical Image Generation and Classification Using Forward Forward Algorithm.
Here, you will find implementations of:
- Forward-Forward Algorithm (FFA): Training with positive and negative passes for energy efficiency.
- Multi-Teacher Knowledge Distillation (MTKD): Distilling knowledge from ResNet-18 (Teacher1) and a smaller CNN (Teacher2) into a Lean Student model.
- Iterative Magnitude Pruning: Techniques to reduce model size and improve efficiency while maintaining performance.
- Experiments on Medical Image Datasets: Code for ODIR-5K (Ocular Disease Recognition) and HAM10000 (Skin Lesion Classification) datasets.
- Evaluation Metrics: Calculation of F1 Score, AUC-ROC, and FID score for performance evaluation.
- Code to Generate Figures: Scripts to reproduce Figures 2 and 3 from the paper, showcasing comparative analysis of algorithms, MTKD evaluation, and pruning effects.
This repository aims to provide a clear and reproducible codebase for researchers and practitioners interested in energy-efficient deep learning for medical image analysis.
This project is licensed under the MIT License - see the LICENSE file for details.
If you use this repository or the ELMAGIC paper in your research or project, please cite it as follows:
@inproceedings{barua2024elmagic,
title={ELMAGIC: energy-efficient lean model for reliable medical image generation and classification using forward forward algorithm},
author={Barua, Saikat and Rahman, Mostafizur and Saad, Mezbah Uddin and Islam, Rafiul and Sadek, Md Jafor},
booktitle={2024 IEEE 3rd International Conference on Computing and Machine Intelligence (ICMI)},
pages={1--5},
year={2024},
organization={IEEE}
}