Insights into the Application of Deep Reinforcement Learning in Healthcare and Materials Science

Reinforcement learning (RL) is a type of machine learning designed to optimize sequential decision-making. While controlled environments have served as a foundation for RL research, due to the growth in data volumes and deep learning methods, it is now increasingly being applied to real-world problems. In our work, we explore and attempt to overcome challenges that occur when applying RL to solve problems in healthcare and materials science.

First, we explore how issues in bias and data completeness affect healthcare applications of RL. To understand how bias has already been considered in this area, we survey the literature for existing methods of bias mitigation related to healthcare data and RL. Based on this review, we conduct our own study where RL is used to learn heparin treatment plans from specific patient populations to demonstrate how healthcare data can affect RL policies and cause bias. Additionally, because synthetic samples are often used to improve the utility of incomplete datasets, we evaluate the impact of data completeness in both panel and longitudinal medical data on the ability of deep generative models to generate similar synthetic samples to augment existing medical datasets.

Next, we focus on materials science applications where RL can potentially be used to automate certain processes. In scanning probe microscopy (SPM), electrical charges applied near ferroelectric domain walls cause the domain wall structure to change. In order to apply RL to control domain walls, we develop a dynamics model trained with SPM images and known physical properties to predict domain wall displacements following an electrical pulse. This dynamics model is then integrated into an environment used to train RL agents that learn how to selectively apply charges that alter domain walls towards a particular structure.