Paper-Conference

R2CC: Reliable and Resilient Collective Communication for Distributed Learning

A fault-tolerant, NCCL-compatible collective communication library that keeps LLM training and serving alive under NIC/link failures with <1.1% training / <3% inference overhead.

wei-wang

• Apr 1, 2027 • 1 min read

ML for Systems

Tidal: Tackling Concept Drift in Provenance-Based Advanced Persistent Threats Detection

A multi-head Transformer-based PIDS that tackles concept drift in APT detection via transfer learning — 27% higher recall and 31% higher precision with half the training data.

yajie-zhou

• Jan 1, 2026 • 1 min read

ML for Systems

Towards Interactive Research Agents for Internet Incident Investigation

LLM-based research agents that simulate experienced researchers for Internet incident investigation.

yajie-zhou

• Nov 1, 2023 • 1 min read