Summary of A Bayesian Approach to Online Planning, by Nir Greshler et al.
A Bayesian Approach to Online Planning
by Nir Greshler, David Ben Eli, Carmel Rabinovitz, Gabi Guetta, Liran Gispan, Guy Zohar, Aviv Tamar
First submitted to arxiv on: 4 Jun 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 The paper combines Monte Carlo tree search and neural networks to improve online planning. A Bayesian approach is developed to quantify uncertainty in neural network outputs, building on classical meta-reasoning ideas. The algorithm uses Thompson sampling for searching action trees, with a finite-time Bayesian regret bound proved. An efficient implementation is proposed for restricted posterior distributions. Additionally, the Bayes-UCB method is applied to trees. Experiments show that when uncertainty estimates are accurate but neural network outputs are inaccurate, the Bayesian approach searches trees more effectively on ProcGen Maze and Leaper environments. The paper also investigates popular uncertainty estimation methods’ accuracy in yielding planning gains. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper improves online planning by using a special combination of two techniques: Monte Carlo tree search and neural networks. It asks if we can make better decisions by thinking about how good or bad the predictions are from these neural networks. A new way to do this is proposed, which uses old ideas from the 1990s. This new approach helps to choose the best actions in situations where the neural network outputs might not be perfect. The paper shows that this approach works better than others on special types of problems. |
Keywords
» Artificial intelligence » Neural network
