Exploration via Epistemic Value Estimation

How to efficiently explore in reinforcement learning is an open problem. Many exploration algorithms employ the epistemic uncertainty of their own value predictions -- for instance compute bonus or upper confidence bound. Unfortunately required difficult estimate general with function approximation. We propose estimation (EVE): a recipe that compatible sequential decision making and neural network approximators. It equips agents tractable posterior over all parameters from which can be computed efficiently. use derive Q-Learning agent observe competitive performance on series benchmarks. Experiments confirm EVE facilitates efficient hard tasks.