MLFCIL: A Multi-Level Forgetting Mitigation Framework for Federated Class-Incremental Learning in LEO Satellites

This feed aggregates recent arXiv papers spanning networking, wireless systems, and AI agent architectures with several security-relevant contributions and dual‑use concerns. Notable items: AI-Sinkhole — an AI-agent augmented DNS discovery/classification + Pi‑Hole blocking framework (with published code) enabling temporary, network-wide blocking of emerging LLM/chatbot services (explainable classifiers using quantized LLMs); TRACE — traceroute-latency based route-change detection (endpoint-only monitoring for routing instability); SegSketch — subnet-aware segmented cardinality estimation to 개선