AutoSND: From Execution Evidence to Structural Policies for Automated Network Dismantling Heuristic Discovery
AutoSND is a three-stage tree search framework for automated network dismantling heuristic discovery. Stage I explores simple heuristics and archives execution evidence. Stage II compiles candidate records into structural policies concerning local signals, neighborhood access, and state update ranges. Stage III continues tree search conditioned on these policies to obtain quality-prioritized (AutoSND-Q) and speed-prioritized (AutoSND-S) candidates. Experiments on 12 real-world networks and 3 large real-world networks show better search performance, stability, and more competitive, structurally interpretable programs.
AutoSND introduces a three-stage tree search framework that automatically discovers network dismantling heuristics by converting execution evidence into structural policies. It outperforms existing methods on real-world networks, producing both quality- and speed-optimized programs with improved interpretability.
The framework's key innovation is the transformation of execution evidence into structural policies that guide subsequent heuristic generation, moving beyond simple candidate screening. This allows the search to learn from failure states and candidate quality, leading to more effective and interpretable dismantling strategies.
Automated heuristic discovery for network dismantling can enhance robustness analysis in critical infrastructure, cybersecurity, and logistics. AutoSND's ability to produce both quality- and speed-optimized solutions makes it adaptable to different operational constraints.
Organizations managing complex networks can use AutoSND to automatically identify critical nodes for targeted protection or disruption, reducing manual effort and improving decision-making in network security and resilience planning.
Next signals include application to dynamic or temporal networks, integration with real-time monitoring systems, and extension to other combinatorial optimization problems where heuristic design is challenging.