Probabilistic DAG Search

TL;DR

This paper introduces a probabilistic framework for search algorithms that leverages latent structures in the search space, improving performance in decision-making tasks like Tic-Tac-Toe and feature selection.

Contribution

It develops a novel probabilistic approach combining Gaussian models and abstractions to better utilize the latent structure of search spaces.

Findings

Outperforms existing non-probabilistic methods in Tic-Tac-Toe.

Shows advantages in feature selection tasks.

Demonstrates the effectiveness of probabilistic modeling in search algorithms.

Abstract

Exciting contemporary machine learning problems have recently been phrased in the classic formalism of tree search -- most famously, the game of Go. Interestingly, the state-space underlying these sequential decision-making problems often posses a more general latent structure than can be captured by a tree. In this work, we develop a probabilistic framework to exploit a search space's latent structure and thereby share information across the search tree. The method is based on a combination of approximate inference in jointly Gaussian models for the explored part of the problem, and an abstraction for the unexplored part that imposes a reduction of complexity ad hoc. We empirically find our algorithm to compare favorably to existing non-probabilistic alternatives in Tic-Tac-Toe and a feature selection application.