Core Concepts
Designing learning-based algorithms for graph searching problems with predictions to optimize traversal efficiency.
Abstract
This content delves into the analysis of learning-based algorithms for graph searching problems with predictions. It explores the challenges of traversing unknown graphs to find hidden goal nodes while minimizing distance traveled. The study focuses on establishing formal guarantees on performance, considering different error models, and proposing algorithms for exploration and planning scenarios. Results show robustness to adversarial errors and optimal or nearly-optimal performance in various settings.
Stats
Banerjee et al. (2023) introduced the problem of graph searching with predictions.
Algorithms are designed for search tasks on unknown weighted graphs.
Guarantees are provided on algorithm performance under different prediction error models.
Lower bounds demonstrate optimality or near-optimality of proposed algorithms.
Numerical experiments show robustness to adversarial errors and good performance in stochastic instances.