Anomalous Sample Detection for AI Security

Deep learning-based AI systems have dominated several long-standing machine learning tasks. Deep neural networks are increasingly used in high-stakes applications, raising worries about their safety and reliability. Fighting anomalous inputs...

Monocular Human Motion Analysis

Our long-term goal of human motion analysis is to help computer learn knowledge of human motion, e.g. how human moves in each action and the limitation on joints freedom. In...