Extracting High-Quality Features From Biomedical Datasets Using Multimodal Autoencoders

This project explores the use of a multimodal autoencoder (implemented in Keras) for learning a shared data representation from the METABRIC breast cancer dataset in order to improve classification accuracy of cancer subtype classifiers by learning high quality features.

Report

The original report can be found here.

Files

main.py - includes code for data pre-processing, and multimodal training+construction.