Abstract | ||
---|---|---|
We revisit the Bellman optimality equation with Nesterovu0027s smoothing technique and provide a unique saddle-point optimization perspective of the policy optimization problem in reinforcement learning based on Fenchel duality. A new reinforcement learning algorithm, called Smoothed Dual Embedding Control or SDEC, is derived to solve the saddle-point reformulation with arbitrary learnable function approximator. The algorithm bypasses the policy evaluation step in the policy optimization from a principled scheme and is extensible to integrate with multi-step bootstrapping and eligibility traces. We provide a PAC-learning bound on the number of samples needed from one single off-policy sample path, and also characterize the convergence of the algorithm. Finally, we show the algorithm compares favorably to the state-of-the-art baselines on several benchmark control problems. |
Year | Venue | Field |
---|---|---|
2017 | arXiv: Learning | Convergence (routing),Saddle,Mathematical optimization,Embedding,Bootstrapping,Computer science,Smoothing,Sample path,Optimization problem,Reinforcement learning |
DocType | Volume | Citations |
Journal | abs/1712.10285 | 3 |
PageRank | References | Authors |
0.39 | 0 | 7 |
Name | Order | Citations | PageRank |
---|---|---|---|
Bo Dai | 1 | 230 | 34.71 |
Albert Shaw | 2 | 26 | 2.45 |
Lihong Li | 3 | 670 | 45.28 |
Xiao, Lin | 4 | 918 | 53.00 |
Niao He | 5 | 212 | 16.52 |
Jianshu Chen | 6 | 883 | 52.94 |
Le Song | 7 | 2437 | 159.27 |