Dorsa Sadigh, Stanford University, USA.
Title: Bringing in the Human in the (Reinforcement) Learning Loop
Abstract: Today, I will talk about the challenges of bringing in humans in the learning loop. Specifically, I will discuss how we can be more data-efficient in robotics when learning from humans by actively querying them. However, this is often not practical in RL settings due to the complexity of active learning of large neural models. Instead, I will go over two approaches that address the need for human data in RL settings: First, I will discuss a modular approach that incorporates inductive biases and enables faster adaptation to human partners. Second, I will briefly discuss how active data acquisition can be helpful for RL agents. Finally, I will conclude by showing that bringing in humans in the learning loop can enable us to aim for more interesting objectives beyond reacting. As an example, we can develop RL agents that can act in non-stationary environments or influence human partners toward more desirable outcomes.
Bio: Dorsa Sadigh is an assistant professor in Computer Science and Electrical Engineering at Stanford University. Her research interests lie in the intersection of robotics, learning, and control theory. Specifically, she is interested in developing algorithms for safe and adaptive human-robot and multi-agent interaction. Dorsa received her doctoral degree in Electrical Engineering and Computer Sciences (EECS) from UC Berkeley in 2017, and received her bachelor’s degree in EECS from UC Berkeley in 2012. She is recognized by awards such as the NSF CAREER award, the AFOSR Young Investigator award, the IEEE TCCPS early career award, MIT TR35, as well as industry awards such as the JP Morgan, Google, and Amazon faculty research awards.