Mohammad Babaeizadeh

Papers

Total Citations

115

H-Index

About

Mohammad Babaeizadeh is a researcher at the forefront of generative modeling and robot learning, with a focus on enabling machines to understand and predict complex sequences in the physical world. His most influential work, "VideoFlow: A Flow-Based Generative Model for Video" (87 citations), introduced a novel approach to video prediction using normalizing flows, demonstrating how generative models can capture intricate real-world phenomena like physical interactions by modeling sequences of future events. This contribution has been foundational for researchers working on predictive models in computer vision. Babaeizadeh has also made significant strides in self-supervised learning for robotics through his work on "Time Reversal as Self-Supervision" (TRASS, 13 citations), which cleverly uses the concept of time reversal to train manipulation policies without heavy supervision, enabling robots to generalize across varying conditions and objectives. More recently, his research on "INFOrmation Prioritization through EmPOWERment in Visual Model-Based RL" explores how reinforcement learning agents can learn to prioritize functionally relevant aspects of visual observations, pushing toward more efficient and robust model-based RL. Through these contributions, Babaeizadeh has established himself as a key innovator at the intersection of generative AI and robotic manipulation.