Hao Fang

Revisiting Backdoor Attacks on LLMs: A Stealthy and Practical Poisoning Framework via Harmless Inputs

Talk Until You Burn Out: Escalating 3D-LLM Overgeneration via Semantic Manipulation

Grounding Language with Vision: A Conditional Mutual Information Calibrated Decoding Strategy for Reducing Hallucinations in LVLMs

Stealthy Shield Defense: A Conditional Mutual Information-Based Approach against Black-Box Model Inversion Attacks

Going Beyond Feature Similarity: Effective Dataset distillation based on Class-aware Conditional Mutual Information

One Perturbation is Enough: On Generating Universal Adversarial Perturbations against Vision-Language Pre-training Models

CALoR: Towards Comprehensive Model Inversion Defense

MIBench: A Comprehensive Benchmark for Model Inversion Attack and Defense