Open World Semi Supervised

Open-world semi-supervised learning (OWSSL) tackles the challenge of training machine learning models on datasets containing both labeled and unlabeled data, where the unlabeled data may include examples from previously unseen classes. Current research focuses on developing robust algorithms, often employing contrastive learning, teacher-student frameworks, and prompt-based methods, to effectively learn representations for both known and novel classes while mitigating the impact of distribution mismatch between labeled and unlabeled data. These advancements aim to improve the generalization capabilities of models in real-world scenarios where encountering new data is inevitable. The impact of OWSSL extends to various applications, including image classification, node classification, and other domains requiring robust handling of unknown data.