Summary of Learning Interpretable Policies in Hindsight-observable Pomdps Through Partially Supervised Reinforcement Learning, by Michael Lanier et al.
Learning Interpretable Policies in Hindsight-Observable POMDPs through Partially Supervised Reinforcement Learning
by Michael Lanier, Ying Xu, Nathan Jacobs, Chongjie Zhang, Yevgeniy Vorobeychik
First submitted to arxiv on: 14 Feb 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI)
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 Deep reinforcement learning has achieved significant success across various domains like video games, robotics, autonomous driving, and drug discovery. However, traditional methods primarily rely on end-to-end learning from high-dimensional observations like images without explicitly reasoning about the true state. The Partially Supervised Reinforcement Learning (PSRL) framework offers an alternative approach that fuses both supervised and unsupervised learning. PSRL leverages a state estimator to extract semantic state information from high-dimensional observations, which are often fully observable during training. This leads to more interpretable policies that combine state predictions with control. In parallel, it captures an unsupervised latent representation. The dual representations – semantic state and latent state – are then fused and used as inputs to a policy network. This blend provides practitioners a flexible spectrum: from emphasizing supervised state information to integrating richer, latent insights. Experimental results demonstrate that PSRL offers a potent balance, enhancing model interpretability while preserving or outperforming traditional methods in terms of reward and convergence speed. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Reinforcement learning has made huge progress across many areas. But most methods rely on high-dimensional observations without thinking about the true state. A new approach called Partially Supervised Reinforcement Learning (PSRL) combines two types of learning: supervised and unsupervised. PSRL uses a special tool to extract useful information from these observations, which helps create more understandable policies. It also captures hidden patterns in the data. By combining these two approaches, PSRL gives people a flexible way to make decisions. This method works well and can even outperform traditional methods. |
Keywords
* Artificial intelligence * Reinforcement learning * Supervised * Unsupervised
