User:Vaish

From RISE
Jump to: navigation, search

Vaishnavh Nagarajan

Program : Bachelors of Technology

Advisor : Dr. Balaraman Ravindran

Research Interests: Machine Learning Theory, Reinforcement Learning


Research Topic : A KWIK analysis of Learning from Demonstrations

Learning from demonstrations (also called as imitation learning) is the problem of learning an expert's policy through sample trajectories provided by the expert. Consider, for example, sequences of (x,y,z) coordinates and the corresponding actions of the arm of a robot that is placing an object at some target. One way to solve this problem is Inverse Reinforcement Learning where we consider the process as a Markov Decision Process and we determine the unknown state-action rewards on which the expert's policy (demonstration) is optimal. We are specifically interested in considering this solution in a Knows-What-It-Knows framework. That is, we would like to devise an online learning algorithm that allows a self-aware learner to ask for demonstrations on-the-fly but only when it encounters an unseen state where it is unable to generalize its experience.

Useful Links :

Knows What It Knows

Apprenticeship Learning via Inverse Reinforcement Learning

Personal Webpage: Vaishnavh Nagarajan

Course taken: Machine Learning, Social Network Analysis, Memory Based Reasoning in Artificial Intelligence, Probabilistic Graphical Models, Natural Language Processing, Reinforcement Learning

Personal tools
Namespaces
Variants
Actions
Navigation
Toolbox