Research

Deepak Pathak is a PhD student at UC Berkeley, working with Prof. Trevor Darrell and Prof. Alexei Efros on the intersection of machine learning, computer vision and robotics. His research work has been featured in popular press outlets, including MIT Technology Review, New Scientist, Quanta Magazine and Wired. Before joining UC Berkeley, Deepak completed his Bachelors with a Gold Medal in Computer Science and Engineering from IIT Kanpur.

Research Summary

Humans demonstrate remarkable ability to generalize their knowledge and skills to new unseen scenarios. One of the primary reasons is that they are continually learning by active exploration of the environment. This is in sharp contrast to our current machine learning algorithms which are incredibly specific in performing the tasks they are trained for.

Deepak’s research goal is to bridge this gap between the conventional supervised learning and the continually adaptive learning in humans. The focus is to build intelligent systems that can perpetually learn by bootstrapping over their own experience without relying on external human supervision. This involves learning general representations of the sensory input using self-supervision and mapping these representation to motor outputs using intrinsically motivated interaction with the environment.

For more information, please visit his website.