Projects

• NEW ARISE: Auto-Research Agents with Recursive Self-Improvement for Cloud-Native Configuration Generation
• NEW SinkOut: Taming Attention Extrema for Efficient LLM Inference
• NEW LLM Agentic Penetration Testing
• NEW EMBER: TCR–Epitope Modeling from Physical Ensembles of Recognition
• NEW LLM Infrastructure Safety

Past Projects

• OASIS: Attention Sinks and Outliers in Attention Residuals
• FROST: Filtering Reasoning Outliers with Attention for Efficient Reasoning
• BOOST: Mind the Inconspicuous: Revealing the Hidden Weakness in Aligned LLMs' Refusal Boundaries
• CRAFT: Contrastive Reasoning Alignment: Reinforcement Learning from Hidden Representations
• DAPA: Decoupled Alignment for Robust Plug-and-Play Adaptation
• TIDES: Test-time Inference Drift Exploitation via Scaling
• GenoArmory: A Unified Evaluation Framework for Adversarial Attacks on Genomic Foundation Models
• RINGS: Accelerating the NextG Protocols Definition to Code Generation with an Automatic and Secure Verification-Compilation Tool-Chain
• Cloud-Native Security
• Cellular Network Security
• Web Security
• Software-Defined Networking
• Mobile Device Security
• Online Social Network Security
• Data Center Networking
• Internet AS Topology: SidewalkEnds
• Stealthy Denial of Service (DoS) Attack Vulnerability Analysis and Defense
• Routing and Location for Sensor and Ad hoc Networks
• Monitoring and Intrusion Detection and Forensics on High-speed Networks
• Scalable and Accurate Monitoring and Diagnosis of Large Networks and Distributed Systems