ACDC: Unified Alert Correlation for Cyber-Physical Power Systems via Prototype-Guided Differentiable Clustering

ACDC is a graph-free framework that couples multi-view alert representation learning with prototype-guided differentiable clustering to correct temporal-semantic bias in cyber-physical SOC alert streams.

This repository contains the reference implementation accompanying the submitted manuscript. Public benchmark results are reported as means over five random seeds with fixed data.random_seed: 42.

Key features:

• Multi-view alert representation (semantic, temporal, host behavior, frequency)
• Prototype-guided differentiable clustering with Iterative Cross-Attention (ICA)
• Jointly optimized clustering-aware representation learning
• One-command reproducibility verification helper

Results:

Citation: Citation will be added upon publication.