Data Augmentation with Diffusion for Open-Set Semi-Supervised Learning Seonghyun Ban

Previous approaches often downweight instances from irrelevant classes to mitigate the negative impact of class distribution mismatch on model training.