backdoor attacks
Backdoor attacks are a type of adversarial threat where malicious modifications are made to a model during training, allowing it to produce specific harmful outputs when triggered by particular inputs, compromising its safety and reliability.
- Attack by Yourself: Effective and Unnoticeable Multi-Category Graph Backdoor Attacks with Subgraph Triggers Pool
- Backdoor Mitigation via Invertible Pruning Masks
- BackdoorLLM: A Comprehensive Benchmark for Backdoor Attacks and Defenses on Large Language Models
- BadVLA: Towards Backdoor Attacks on Vision-Language-Action Models via Objective-Decoupled Optimization
- Defending Multimodal Backdoored Models by Repulsive Visual Prompt Tuning
- FedRACE: A Hierarchical and Statistical Framework for Robust Federated Learning
- ICLScan: Detecting Backdoors in Black-Box Large Language Models via Targeted In-context Illumination
- LoSplit: Loss-Guided Dynamic Split for Training-Time Defense Against Graph Backdoor Attacks
- MARS: A Malignity-Aware Backdoor Defense in Federated Learning
- RepGuard: Adaptive Feature Decoupling for Robust Backdoor Defense in Large Language Models
- SNEAKDOOR: Stealthy Backdoor Attacks against Distribution Matching-based Dataset Condensation
- Stealthy Yet Effective: Distribution-Preserving Backdoor Attacks on Graph Classification
- Temporal Logic-Based Multi-Vehicle Backdoor Attacks against Offline RL Agents in End-to-end Autonomous Driving
- Virus Infection Attack on LLMs: Your Poisoning Can Spread "VIA" Synthetic Data
- Who Speaks for the Trigger? Dynamic Expert Routing in Backdoored Mixture-of-Experts Transformers