Summary of Rvi-sac: Average Reward Off-policy Deep Reinforcement Learning, by Yukinari Hisaki et al.
RVI-SAC: Average Reward Off-Policy Deep Reinforcement Learning
by Yukinari Hisaki, Isao Ono
First submitted to arxiv on: 4 Aug 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: None
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 proposed off-policy deep reinforcement learning (DRL) method, RVI-SAC, utilizes the average reward criterion as an alternative to the discounted reward criterion. This is particularly relevant for continuing tasks where a discrepancy between training objectives and performance metrics can occur. RVI-SAC is an extension of the state-of-the-art SAC method and consists of three components: critic updates based on RVI Q-learning, actor updates introduced by the average reward soft policy improvement theorem, and automatic adjustment of Reset Cost to apply the average reward reinforcement learning to tasks with termination. The method is evaluated on the Gymnasium’s Mujoco locomotion tasks, demonstrating competitive performance compared to existing methods. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper proposes a new way for computers to learn from experiences without following all the rules. Usually, these computers use a special number called “discounted reward” to decide what actions are best. But this can sometimes cause problems if the task keeps going forever. To solve this, the researchers created a new method called RVI-SAC that uses a different type of reward called “average reward”. This allows the computer to learn better in tasks that never end. The team tested their method on some simple robot tasks and found it worked just as well as other popular methods. |
Keywords
» Artificial intelligence » Reinforcement learning
