Summary of No-regret Reinforcement Learning in Smooth Mdps, by Davide Maran et al.
No-Regret Reinforcement Learning in Smooth MDPs
by Davide Maran, Alberto Maria Metelli, Matteo Papini, Marcello Restell
First submitted to arxiv on: 6 Feb 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary A novel structural assumption called -smoothness is introduced to address open challenges in reinforcement learning (RL) for continuous state and/or action spaces. The authors propose two algorithms, Legendre-Eleanor and Legendre-LSVI, which build upon orthogonal feature maps based on Legendre polynomials to construct MDP representations. While Legendre-Eleanor achieves no-regret guarantees under weaker assumptions but is computationally inefficient, Legendre-LSVI runs in polynomial time for a smaller class of problems. The algorithms are compared to state-of-the-art RL theory results, showing that they achieve the best guarantees. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Reinforcement learning (RL) tries to solve big problems where computers learn from trying different actions and seeing what works best. Right now, it’s hard to get good guarantees when there are lots of possible states and actions. Some smart people have come up with ideas to help with this, but they mostly work for very specific situations. This paper introduces a new way to think about problems that helps make RL better in many different scenarios. Two algorithms are proposed that use special maps to help computers learn from experiences. One algorithm is good at getting the best results but takes a long time, while the other one is faster but not as good. The authors compare their ideas to what other smart people have done and show that they do the best job. |
Keywords
* Artificial intelligence * Reinforcement learning
