MIT EECS — Quanta Computer Undergraduate Research and Innovation Scholar
Learning Video Dynamics With Recurrent Neural Networks
Physical scene understanding requires mastery of two deep concepts: the reasoning about objects in the scene and performing causal inferences on their interactions. My project investigates approaches to disentangle the objects and physical laws of a video by using a recurrent neural network architecture to learn a physics engine. Explicitly binding learned information to variables would allow the program to flexibly manipulate the disentangled components of the scene and gives rise to many interesting applications, such as counterfactual reasoning
I am interested in developing and implementing computational models of intelligence. This past summer I conducted research in deep learning with Professor Honglak Lee at the University of Michigan. I hope to gain a deeper understanding of the math and theory behind machine learning and inference, and I am excited to research at the intersection of both the engineering and philosophy of artificial intelligence.