Summary of A Training Data Recipe to Accelerate A* Search with Language Models, by Devaansh Gupta et al.
A Training Data Recipe to Accelerate AuthorLineProcess.function Search with Language Models
by Devaansh Gupta, Boyang Li
First submitted to arxiv on: 13 Jul 2024
Categories
- Main: Artificial Intelligence (cs.AI)
- Secondary: Machine Learning (cs.LG)
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 Combining Large Language Models (LLMs) with heuristic search algorithms like A* holds the promise of enhanced LLM reasoning and scalable inference. This paper investigates coreset selection for training data of LLM heuristic learning, exploring the interaction between search algorithm and machine learning model. The authors empirically disentangle requirements of A* from those of LLM to generalize on this task. Surprisingly, they find an overlap between their requirements; A* requires accurate predictions near the goal, while LLMs need the same nodes for effective generalization. With these insights, they derive a data-selection distribution for learning LLM-based heuristics. The results show that their technique reduces iterations required to find solutions by up to 15x and speeds up search by up to 5x on three classical planning domains: maze navigation, Sokoban, and sliding tile puzzles. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper combines special language models with a type of computer searching called A* to make the models work better and faster. The researchers studied how to pick the most important data for training these models. They found that both A* and the language model need similar information to work well, which helped them create a new way to select this data. This new method worked much faster than before, reducing the time it took to solve puzzles by up to 15 times. |
Keywords
» Artificial intelligence » Generalization » Inference » Language model » Machine learning
