information leakage
Information leakage occurs when sensitive information is inadvertently exposed in the training process, potentially leading to models that perform better during evaluation but fail to generalize effectively. In AI, preventing leakage is crucial for maintaining model robustness and trustworthiness.
- AgentDAM: Privacy Leakage Evaluation for Autonomous Web Agents
- Approximate Domain Unlearning for Vision-Language Models
- Bits Leaked per Query: Information-Theoretic Bounds for Adversarial Attacks on LLMs
- Enhancing Privacy in Multimodal Federated Learning with Information Theory
- Time Travel is Cheating: Going Live with DeepFund for Real-Time Fund Investment Benchmarking