I am specializing on understanding how imitation learning can be made more efficient in certain environmental settings.
As a long term reserach goal, I am interested in the human inspired learning of artificial agents.
Machine agents are increasingly being designed and built to be used in close association with the humans. Specifically, intelligent machines are heavily used in the automatic tutoring systems that assist humans in learning numerous skills like playing a musical instrument, becoming an expert in a scientific/mathematical subject, flying an aircraft, performing a surgery, etc.
Reinforcement learning or the inverse of it (inverse reinforcement learning) can be used to design the automatic tutoring systems for such sequential decision making tasks. However, machines trained purely on the raw data might not account for the idiosyncrasies associated with the human behaviour. Idiosyncrasies in the human behaviour can range from the sensation of pain or pleasure to inductive biases. In some cases, it can become important to explicitly include these human traits into the learning procedure of the machine so that the machine can learn to devise a human friendly tutoring plan. My research interest stems from this need, and I want to understand when, why and how humans need to be brought into the learning loop of the reinforcement learning agents.