JointDiffusion: Joint representation learning for generative, predictive, and self-explainable AI in healthcare

Joint machine learning models that allow synthesizing and classifying data often offer uneven performance between those tasks or are unstable to train. In this work, we depart from a set of empirical observations that indicate the usefulness of internal representations built by contemporary deep diffusion-based generative models not only for generating but also predicting. We then propose to extend the vanilla diffusion model with a classifier that allows for stable joint end-to-end training with shared parameterization between those objectives. The resulting joint diffusion model outperforms recent state-of-the-art hybrid methods in terms of both classification and generation quality on all evaluated benchmarks. On top of our joint training approach, we present its application to the medical data domain, where we show how joint training can aid with the problems crucial in the medical data domain. We show that our Joint Diffusion achieves superior performance in semi-supervised setup, where human annotation is scarce, while at the same time providing decisions explanations through counterfactual examples generation.