Hello! I am a masters student at Harvard Computational Science. I am working on the application of inverse reinforcement learning and interpretable machine learning in healthcare. My advisor is Finale Doshi-velez.
Prior to Harvard, I enjoyed a fantastic adventure as a co-founder of a profitable hardware accelerator startup and as web analytics engineer at Plivo (yc s12). I graduated with honours from Aalto University in Business Technology and Computer Science where I had a ton of fun working on interesting projects.
My research focuses on the robust, off-policy, sample-efficient inverse reinforcement learning in healthcare. Concretely, I am studying how we can develop efficient and robust algorithms that can extract human-interpretable, general knowledge from expert demonstration. Furthremore, I am interested in developing theoretical bounds on the recovery of expert policy under noisy (suboptimal) demonstrations and a model-free off-policy inverse reinforcement learning method under unkonwn dynamics. Of course, the healthcare domain carry all these interesting problems.
Most recent research projects include:
- Understanding clinician’s Sepsis treatment: how can we apply Inverse Reinforcement Learning to better understand and imitate the behaviors of clinicians?
- Interpretabie and fair loan lending: how can we perform a robust posthoc interpretability and fairness evaluation of a black-box model? (partnering with Square Capital)
Last semester, I co-developed a Texas Holdem Poker playing agent with a Neural Fictitious Self-play algorithm. Last Spring, I also co-implemented a count-based exploration RL agent for the notorious Montezuma’s revenge where I got the first position on OpenAI’s leaderboard.
As a budding researcher, I am about to make my first submission to MLHC.