Summary of Bayesian Inverse Reinforcement Learning For Non-markovian Rewards, by Noah Topper et al.
Bayesian Inverse Reinforcement Learning for Non-Markovian Rewards
by Noah Topper, Alvaro Velasquez, George Atia
First submitted to arxiv on: 20 Jun 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 Inverse reinforcement learning (IRL) aims to infer a reward function from expert behavior, but current approaches are limited to Markovian rewards. We propose Bayesian IRL (BIRL), a framework for inferring Reward Machines (RMs) directly from expert behavior, without access to the reward signal. Our BIRL framework defines a new reward space, adapts expert demonstrations to include history, and proposes novel modifications to simulated annealing for maximizing the reward posterior. We demonstrate that our method performs well in optimizing according to its inferred reward and compares favorably to existing methods learning binary non-Markovian rewards. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Imagine trying to figure out what makes a robot make good decisions. This paper proposes a new way to do this, called Bayesian Inverse Reinforcement Learning (IRL). The old way of doing IRL only works for simple situations where the reward is based on the current state. But in real life, rewards often depend on more than just the current state. Our new approach can handle these more complex situations and even figure out what makes a robot make good decisions without knowing what the reward is. |
Keywords
» Artificial intelligence » Reinforcement learning
