DiL: Discrete-anchored Representation Alignment for Semi-Supervised Continual Learning

Leveraging the unlabeled stream is crucial yet challenging in Semi-Supervised Continual Learning (SSCL) under continual class expansion. Existing SSCL methods typically enforce dense pseudo-label consistency and indiscriminate distillation on unlabeled data, which can reinforce errors and intensify base–novel interference. To address these issues, we propose Discrete-anchored Incremental Learning (DiL) to ground continual updates on reliable discrete anchors that remain stable under noisy pseudo-labels. DiL introduces Discrete Contrastive Distillation (DCD), which discretizes the distillation pathway and performs anchor-referenced selective distillation to curb error reinforcement. Meanwhile, Class-Aware Channel-Chunked Encoding (CACE) learns channel-chunked representations and exploits the confusion matrix induced by the discrete anchors to separate novel from confusable base classes. Extensive experiments on multiple datasets show that DiL achieves state-of-the-art performance across diverse SSCL protocols.

Machine learning systems often struggle to learn new categories over time without forgetting previously learned knowledge, especially when only a small amount of labeled data is available. This challenge makes it difficult for learning systems to adapt to real-world environments where data continuously changes. In this work, we propose a new learning approach that helps models organize knowledge using stable category references. These references guide the model when learning new information, reducing confusion between similar categories while preserving earlier knowledge. Our method also improves the model’s ability to distinguish newly learned categories from older ones. We evaluate our approach on several challenging learning benchmarks and show that it consistently improves performance, particularly in difficult settings with very limited labeled data. These results suggest that our method can help build more reliable and adaptable AI systems for dynamic real-world applications.