Summary of Multi Task Inverse Reinforcement Learning For Common Sense Reward, by Neta Glazer et al.
Multi Task Inverse Reinforcement Learning for Common Sense Reward
by Neta Glazer, Aviv Navon, Aviv Shamsian, Ethan Fetaya
First submitted to arxiv on: 17 Feb 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 This research paper proposes a novel approach to addressing the challenge of providing a detailed reward function for agents in complex real-world environments. The authors argue that misalignment between the reward and desired behavior can lead to unintended outcomes, such as “reward hacking.” To mitigate this issue, they suggest disentangling the reward into two parts: task-specific rewards outlining the particulars of the task, and unknown common-sense rewards indicating expected agent behavior within the environment. The study explores how this common-sense reward can be learned from expert demonstrations, highlighting the limitations of inverse reinforcement learning in training agents with a useful reward function. Instead, the authors propose using multi-task inverse reinforcement learning to learn a useful reward function that can be applied across multiple tasks. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is all about helping machines learn how to behave properly in real-world situations. Right now, it’s hard to teach machines what behavior is good and what isn’t because we don’t have the right “reward” system in place. This reward system tells the machine what it did well or poorly. But if we get it wrong, the machine might do things that aren’t what we wanted! To fix this, the researchers suggest breaking down the reward into two parts: one that says what to do exactly and another that teaches good behavior. They also show that learning from expert demonstrations is not enough to get a good reward function. Instead, they propose teaching machines multiple tasks at once to learn how to behave well in different situations. |
Keywords
* Artificial intelligence * Multi task * Reinforcement learning
