RnD Project

The project is supervised by Prof. Kshitij Jadhav and Prof. Ganesh Ramakrishnan in DH307 offered by the KCDH dept. at IIT-Bombay. Details about the project are listed below:

Diffusion Models for medical data Augmentation in Federated learning setting for preserving privacy

Abstract:

Diffusion models can be utilized for medical data augmentation in a federated learning setting to preserve privacy. Federated learning allows training machine learning models across multiple institutions without directly sharing sensitive patient data. Here's an overview of the approach

1. Data Privacy and Federated Learning: In a federated learning setting, data remains decentralized and resides within different medical institutions or devices. Privacy is maintained by keeping the data local and performing model training and updates on the local data without exchanging the raw data itself.
2. Diffusion Models: Diffusion models, such as generative models like Variational Autoencoders (VAEs) or Generative Adversarial Networks (GANs), can be employed for medical data augmentation. These models learn the underlying distribution of the medical data and generate new synthetic samples that preserve the statistical properties of the original data.
3. Local Data Augmentation: At each local medical institution or device participating in the federated learning framework, diffusion models can be trained on the local data. These local models learn the local data distribution

Dataset

Link to the preprocessed dataset: https://drive.google.com/drive/u/0/folders/1gxOiEOzuxtfKkC1oslO9likQY2l_ysBh Code to pre-process the data will soon be pushed to this repo

Link to the raw dataset: https://drive.google.com/file/d/1LHxvqJaD23CEF4QBCTmV9qYo-FDusnJn/view

How to run?

First of all, clone the popular wGAN repo as:

git clone https://github.com/martinarjovsky/WassersteinGAN/blob/master/main.py

Add the image dataset folder to this and run the following command, to know more about the flags, see the generate.py file!

Old Command to run:

python3 main.py --dataset folder --dataroot Dataset --cuda --workers 20 --niter 10000

New Command to run:

python3 main.py --datasetA folder --datasetB folder --datarootA Dataset-A --datarootB Dataset-B --cuda --workers 20 --imageSize 128 --niter1 4000 --niter2 10000 --lrD 1e-5 --lrG 1e-5 --num_classes 2 --num_clients 2 --adam 2>&1 | tee training_log.txt