Summary of Dynamical-vae-based Hindsight to Learn the Causal Dynamics Of Factored-pomdps, by Chao Han et al.
Dynamical-VAE-based Hindsight to Learn the Causal Dynamics of Factored-POMDPs
by Chao Han, Debabrota Basu, Michael Mangan, Eleni Vasilaki, Aditya Gilra
First submitted to arxiv on: 12 Nov 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Machine Learning (stat.ML)
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 paper tackles the challenge of learning environmental dynamics from partial observations in Partially Observable Markov Decision Processes (POMDPs). Typically, POMDP state representations are inferred from past observations and actions. However, incorporating future information is crucial to accurately capture causal dynamics and improve state representations. To address this, the authors introduce a Dynamical Variational Auto-Encoder (DVAE) that learns causal Markovian dynamics from offline trajectories in a POMDP setting. The method employs an extended hindsight framework, integrating past, current, and multi-step future information within a factored-POMDP setting. Experimental results show that this approach effectively uncovers the causal graph governing hidden state transitions, outperforming history-based and typical hindsight-based models. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about finding patterns in complex systems where we can’t see everything at once. In these situations, it’s hard to figure out what’s really going on. The authors developed a new way to learn from past observations and future predictions to create better maps of the system. They call this method Dynamical Variational Auto-Encoder (DVAE). It helps us understand how things change over time and makes more accurate predictions about what might happen next. |
Keywords
* Artificial intelligence * Encoder
