Compact-A∗: Space-Efficient Fixed-Length Path Optimization

ICLR 2026 Conference Submission16186 Authors

Keywords: path optimization, decoding, HMMs

TL;DR: We introduce space-efficient A*-based algorithms for fixed-length path search.

Abstract: Several optimization problems seek a path in a state space to minimize a cost function under a length constraint. Traditionally, these are solved by A∗ search or dynamic programming (DP), as in Viterbi decoding for Hidden Markov Models. In all cases, solutions require memory commensurate with a search space that grows linearly in both state space size and path length. In this paper, we propose compact-A∗, a framework that limits the growth of the A∗ priority queue to determine the minimum cost for a path of predetermined length in a space-efficient manner and then constructs such a path by a divide-and-conquer strategy that eliminates the memory overhead. We apply compact-A∗ to Viterbi decoding and further highlight its generality with an application to V-optimal histogram construction. Our experimental results demonstrate significant improvements over state-of-the-art solutions in runtime and memory consumption.

Primary Area: optimization

Submission Number: 16186