AutoSND: Automated Network Dismantling Heuristic Discovery via Tree Search

A new framework named AutoSND has been developed by researchers to automate the discovery of heuristics for network dismantling through a three-stage tree search process. This issue is crucial for assessing the robustness and vulnerability of intricate systems, yet effective heuristics typically need to balance computational efficiency and effectiveness, often requiring manual design. While existing methods utilizing large language models can generate and evaluate candidates, they face challenges in converting candidate quality or execution failures into structural guidance for future generations. AutoSND overcomes this by initially exploring simple heuristics and collecting execution data, then creating structural policies based on local signals, neighborhood access, and state update ranges, and finally conducting a tree search based on these policies to identify candidates prioritized by quality and speed. This framework is elaborated in a paper available on arXiv (2608.03653), categorized as an announcement type 'new', marking progress in the automation of network dismantling algorithm design and enhancing the analysis of complex systems.