A Reduction from Reinforcement Learning to No-Regret Online Learning
Cheng, Ching-An, Combes, Remi Tachet des, Boots, Byron, Gordon, Geoff
We present a reduction from reinforcement learning (RL) to no-regret online learning based on the saddle-point formulation of RL, by which "any" online algorithm with sublinear regret can generate policies with provable performance guarantees. This new perspective decouples the RL problem into two parts: regret minimization and function approximation. The first part admits a standard online-learning analysis, and the second part can be quantified independently of the learning algorithm. Therefore, the proposed reduction can be used as a tool to systematically design new RL algorithms. We demonstrate this idea by devising a simple RL algorithm based on mirror descent and the generative-model oracle. For any $\gamma$-discounted tabular RL problem, with probability at least $1-\delta$, it learns an $\epsilon$-optimal policy using at most $\tilde{O}\left(\frac{|\mathcal{S}||\mathcal{A}|\log(\frac{1}{\delta})}{(1-\gamma)^4\epsilon^2}\right)$ samples. Furthermore, this algorithm admits a direct extension to linearly parameterized function approximators for large-scale applications, with computation and sample complexities independent of $|\mathcal{S}|$,$|\mathcal{A}|$, though at the cost of potential approximation bias.
Nov-13-2019
- Country:
- North America
- United States
- Iowa (0.04)
- Illinois > Cook County
- Chicago (0.04)
- Canada > Quebec
- Montreal (0.04)
- United States
- Europe > United Kingdom
- England
- Greater London > London (0.04)
- Cambridgeshire > Cambridge (0.04)
- England
- North America
- Genre:
- Research Report (0.50)
- Industry:
- Education > Educational Setting > Online (0.82)
- Technology: