PhD Candidate
Massachusetts Institute of Technology
Areas of Interest
- Artificial Intelligence
- Control, Intelligent Systems, and Robotics
- Programming Systems
Poster
Program Synthesis for Robust Robots
Abstract
Deep learning has successfully solved many challenging problems in artificial intelligence. However, deep learning has several limitations that affect its usability–for eg., it is data-hungry, produces opaque and brittle models, and lacks any guarantees. My research alleviates these limitations by proposing a neurosymbolic approach to learning that combines traditional machine learning with program synthesis—a field dedicated to automatically generating programs from specifications. Programs are known to have rich symbolic structures that are interpretable, generalizable and robust; and my work investigates how learning program models can contribute to better intelligent systems. The neurosymbolic learning approach has a wide range of applications; in particular, my focus is robotics because this is a domain where reliability, robustness and ability to enforce constraints are essential for safety.
Bio
I am a Ph.D. candidate at MIT advised by Armando Solar-Lezama. I work across artificial intelligence, program synthesis, and robotics. My work has appeared in machine learning venues (NeurIPS, ICLR), robotics venues (ICRA) and programming languages venues (POPL, TACAS, SAT).
