AI Safety

Learning with Safety Constraints: Sample Complexity of Reinforcement Learning for Constrained MDPs AAAI 2021 LiBRe: A Practical Bayesian Approach to Adversarial Detection CVPR 2021 Limitations of Post-Hoc Feature Alignment for Robustness CVPR 2021 LIRA: Learnable, Imperceptible and Robust Backdoor Attacks ICCV 2021 Look Before You Leap: Safe Model-Based Reinforcement Learning with Human Intervention CoRL 2021 Lyapunov-stable neural-network control RSS 2021 Manipulating SGD with Data Ordering Attacks NeurIPS 2021 Misspecification in Prediction Problems and Robustness via Improper Learning AISTATS 2021 Mitigating Political Bias in Language Models through Reinforced Calibration AAAI 2021 Model-free Safe Control for Zero-Violation Reinforcement Learning CoRL 2021 Model-Targeted Poisoning Attacks with Provable Convergence ICML 2021 Moral Stories: Situated Reasoning about Norms, Intents, Actions, and their Consequences EMNLP 2021 Morié Attack (MA): A New Potential Risk of Screen Photos NeurIPS 2021 Multi-Expert Adversarial Attack Detection in Person Re-Identification Using Context Inconsistency ICCV 2021 Multi-granularity Textual Adversarial Attack with Behavior Cloning EMNLP 2021 Multi-Objective Reinforcement Learning for Designing Ethical Environments IJCAI 2021 Neural Architecture Dilation for Adversarial Robustness NeurIPS 2021 Neural Attention Distillation: Erasing Backdoor Triggers from Deep Neural Networks ICLR 2021 Neural Lyapunov Redesign L4DC 2021 On exploration requirements for learning safety constraints L4DC 2021 ONION: A Simple and Effective Defense Against Textual Backdoor Attacks EMNLP 2021 Optimal Policies Tend To Seek Power NeurIPS 2021 Pairing Weak with Strong: Twin Models for Defending Against Adversarial Attack on Speaker Verification Interspeech 2021 PointGuard: Provably Robust 3D Point Cloud Classification CVPR 2021 Poisoning Knowledge Graph Embeddings via Relation Inference Patterns ACL 2021