Open Set Scenario

Open-set recognition (OSR) addresses the challenge of machine learning systems handling data containing unknown classes, unlike the traditional closed-set assumption. Current research focuses on developing algorithms and model architectures that can reliably classify known classes while accurately rejecting or identifying unknown data, employing techniques like distance-based methods, multi-task learning, and prompt tuning. This field is crucial for building robust and safe AI systems capable of operating in real-world environments where encountering novel or unexpected data is inevitable, impacting diverse applications from robotics and malware detection to speech synthesis and knowledge graph completion.