Bhavik V. Nagda

Bhavik V. Nagda
MIT EECS | CS+HASS Undergraduate Research and Innovation Scholar
Advisor: Joshua Tenenbaum
Department: EECS
Years: 2019-2020
Research Project Title:

Inductive Human Bias Via Human-Based Learning

abstract:Our cognitive ability to learn complex behaviors from little information and data is both familiar and elusive to researchers. From relatively few examples, our brains can accurately learn and compose remarkable tasks; current state-of-the-art machines are far from achieving this phenomenon. A potential explanation for this discrepancy is the theory of developmental start-up software: humans are born with intuitive priors of physics and psychology. The focus of this study is to identify few-shot learning methods for Atari games by infusing human priors in the agent’s understanding of the game state. We suspect that such ‘start-up’ software embeddings pave a stronger path forward toward universal artificial intelligence.

"This SuperUROP project is my first foray into probabilistic artificial intelligence techniques, and I am beyond excited to apply the probabilistic analysis skills from inference class and my software engineering skills to build out a more intelligent agent. As a student broadly interested in graduate school, SuperUROP is the perfect opportunity to explore a future in academia or research."