Summary of Deflated Dynamics Value Iteration, by Jongmin Lee et al.
Deflated Dynamics Value Iteration
by Jongmin Lee, Amin Rakhsha, Ernest K. Ryu, Amir-massoud Farahmand
First submitted to arxiv on: 15 Jul 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Optimization and Control (math.OC)
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 The Value Iteration algorithm is a fundamental procedure in reinforcement learning that computes the value function of a Markov decision process. However, its error convergence rate slows down significantly when the discount factor is close to 1. To address this issue, we propose Deflated Dynamics Value Iteration (DDVI), which leverages matrix splitting and deflation techniques to effectively remove the top dominant eigen-structure of the transition matrix. This leads to a faster convergence rate of O(γ^k |λ_s+1|^k). We further extend DDVI to the reinforcement learning setting, introducing Deflated Dynamics Temporal Difference (DDTD) algorithm. Our empirical results demonstrate the effectiveness of these proposed algorithms. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The Value Iteration algorithm is used in many areas of machine learning. It helps calculate the value function of a Markov decision process. However, when the discount factor is close to 1, it gets slower. We developed an improved version called Deflated Dynamics Value Iteration (DDVI). This new method uses special techniques to remove some parts of the transition matrix. This makes it run faster. We also extended this idea to a related area called reinforcement learning. Our tests showed that these new algorithms work well. |
Keywords
» Artificial intelligence » Machine learning » Reinforcement learning
