Summary of Configurable Mirror Descent: Towards a Unification Of Decision Making, by Pengdeng Li et al.
Configurable Mirror Descent: Towards a Unification of Decision Making
by Pengdeng Li, Shuxin Li, Chang Yang, Xinrun Wang, Shuyue Hu, Xiao Huang, Hau Chan, Bo An
First submitted to arxiv on: 20 May 2024
Categories
- Main: Artificial Intelligence (cs.AI)
- Secondary: Computer Science and Game Theory (cs.GT); Machine Learning (cs.LG); Multiagent Systems (cs.MA)
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 This paper aims to develop a single algorithm that can tackle all types of decision-making problems, including single-agent, cooperative multi-agent, competitive multi-agent, and mixed cooperative and competitive scenarios. To achieve this goal, the authors propose three main contributions: generalized mirror descent (GMD), configurable mirror descent (CMD), and GameBench, a comprehensive benchmark covering 15 academic-friendly games across different decision-making categories. The GMD algorithm generalizes multiple historical policies and works with a broader class of Bregman divergences. The CMD algorithm introduces a meta-controller to dynamically adjust hyper-parameters in GMD conditional on evaluation measures. Extensive experiments demonstrate that CMD achieves empirically competitive or better outcomes compared to baselines while providing the capability of exploring diverse dimensions of decision making. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper wants to make computers smart enough to make good decisions in all kinds of situations. Right now, different types of decision-making problems are solved using separate methods. The authors ask: Can we have one algorithm that works for everything? They propose three ideas to help answer this question. First, they suggest an algorithm called GMD that can learn from past experiences and work with many different types of “distance” measures. Second, they introduce a new way to adjust the settings of GMD based on how well it’s doing. Third, they create a big test set called GameBench that includes 15 games covering various decision-making scenarios. The authors show that their approach can solve problems as well or better than existing methods. |
