Wei Zhou

The paper introduces REFORGE, a black-box red-teaming framework that uses adversarial image prompts to reveal persistent vulnerabilities in current Image Generation Model Unlearning (IGMU) methods.

This paper provides the first comprehensive, end-to-end survey dedicated to the security of Retrieval-Augmented Generation (RAG) systems, systematically mapping threats, defenses, and benchmarks across the entire pipeline.

The paper proposes CAAP, a capture-aware adversarial patch framework, demonstrating that deep palmprint recognition systems remain vulnerable to physically realizable attacks despite existing defenses.

The paper introduces FlowSteer, a prompt-only attack that exploits vulnerabilities in how multi-agent LLM systems plan workflows, significantly increasing the success rate of malicious signal propagation.

The paper introduces FIDO, a novel framework that significantly boosts firmware fuzzing efficiency by accurately managing the timing and quantity of input delivery based on the firmware's internal input availability checks.

The paper introduces ADR, a novel, production-proven detection system that provides high-fidelity security monitoring for AI agents operating via the Model Context Protocol, significantly outperforming existing state-of-the-art baselines.

The paper proposes TCP-MCP, a co-evolution framework that jointly optimizes agent prompts and communication topologies to design highly efficient and effective multi-agent systems.

This paper introduces Capability Self-Assessment (CSA), a crucial ability for LLMs to recognize their limitations, and demonstrates that reinforcement learning is an effective method for teaching this skill without degrading the model's core capabilities.

ProtoAda introduces a prototype-guided, format-aware adaptive tuning framework to improve multimodal continual instruction tuning by ensuring task assignment and parameter updates respect heterogeneous output structures.

The paper proposes a novel nonparametric mutual information estimator to robustly quantify dependence between heterogeneous temporal data, specifically continuous time series and discrete event sequences.

CRAM proposes a novel framework for Multimodal Continual Instruction Tuning that balances task isolation and parameter efficiency by using centroid-guided routing and adaptive MoE to prevent catastrophic forgetting.

GLOVES is a flow-based adaptation method that selectively corrects non-expert robot actions by guiding them toward a task-specific expert action distribution, thereby improving performance while maintaining agent autonomy.