A way around the exploration-exploitation dilemma
Erik J Peterson, Timothy Verstynen · 2019 Conference on Cognitive Computational Neuroscience · 2019
The exploration-exploitation dilemma is a fundamental but intractable problem in the learning and decision sciences. In this problem the goal of exploration and exploitation is to maximize reward. Here we challenge this basic form. We conjecture the dilemma can be viewed as a competition between exploiting known rewards or exploring to learn a world model, a simplified concept of memory borrowed from computer science. We prove this competition is tractable. It can be solved by a simple greedy algorithm. And this solution has the properties expected of an optimal solution to original dilemma--finding maximum value with no regret.