Summary of Beyond A*: Better Planning with Transformers Via Search Dynamics Bootstrapping, by Lucas Lehnert et al.
Beyond A*: Better Planning with Transformers via Search Dynamics Bootstrapping
by Lucas Lehnert, Sainbayar Sukhbaatar, DiJia Su, Qinqing Zheng, Paul Mcvay, Michael Rabbat, Yuandong Tian
First submitted to arxiv on: 21 Feb 2024
Categories
- Main: Artificial Intelligence (cs.AI)
- 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 Transformers have achieved significant advancements in various applications, yet they still lag behind traditional symbolic planners for solving complex decision-making tasks. Our work demonstrates how to train Transformers to solve these complex planning tasks by training an encoder-decoder Transformer model to predict the search dynamics of the A^* search algorithm. We fine-tuned this model to obtain a Searchformer, which optimally solves previously unseen Sokoban puzzles 93.7% of the time using up to 26.8% fewer search steps than the initial implementation used for training. Our method expresses A^*’s search dynamics as a token sequence outlining when task states are added and removed into the search tree during symbolic planning. Searchformer significantly outperforms baselines that predict the optimal plan directly, with a 5-10 times smaller model size and a 10 times smaller training dataset. We also demonstrate how Searchformer scales to larger and more complex decision-making tasks with improved percentages of solved tasks and shortened search dynamics. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Searchformers are special types of Transformers that can solve big planning problems. They work by learning how to predict the steps needed to solve a problem, kind of like following a recipe. This helps them make good decisions even when they haven’t seen the problem before. The Searchformer model is really good at solving Sokoban puzzles, which are a type of puzzle that involves moving boxes around. It can even do it faster and better than other models! This is important because it shows that Searchformers can be used to solve lots of different types of problems. |
Keywords
» Artificial intelligence » Encoder decoder » Token » Transformer
