Coaching How to Search

Searching is commonly addressed by either directly programming a machine to do so or letting it explore the search state space and craft a path “on its own”, as in Reinforcement Learning. Both approaches place most of the computational and / or cognitive load to one end, be it the programmer or the machine learner. An alternative way, balancing load between the two learning ends, is to have a human or machine coach possesing explicit knowledge on the underlying search problem to provide adequate advice upon a machine learner’s request for further knowledge. In this paper, building on previous work on Machine Coaching, specialized in the particular context of searching, a novel learning paradigm is presented, investigating its efficacy and conformity. Potential next steps and real-world applications of the proposed paradigm are also discussed.