Research Project Title:
Deep Learning Approaches to Estimate Treatment Effects under Dynamic, Time-Varying Treatment Regimes
abstract:Estimating the effects of different treatment strategies becomes difficult without deep learning due to high-dimensional patient histories and dynamic, time-varying treatment options. My project entails exploring different deep learning models to develop a sequential modeling framework for personalized patient outcome predictions. The hope is to apply this model to estimate the effects of various treatment strategies in an intensive care unit (ICU) setting. In particular, I will explore and implement methods designed for time-series data, such as TimeGAN and WaveNet. Time permitting, I plan on exploring domain adaptation, which is looking at how to apply a well-performing model to a different dataset and distribution than the original dataset.
Since doing a UROP last year, which focused on applying machine learning (ML) to solar cell research, I have been interested in using ML and optimization to make different processes more efficient. Through this SuperUROP, I hope to gain knowledge and experience in applying ML algorithms to the medical field and work on improving the state-of-art models. I am most excited to learn new technical concepts and have presentable results at the end!