Online Class Incremental Learning

Online class incremental learning (OCIL) focuses on training machine learning models that can continuously learn new classes from a data stream without access to past data, addressing the "catastrophic forgetting" problem where previously learned knowledge is lost. Current research emphasizes developing algorithms that mitigate forgetting, often employing techniques like generative models, self-distillation, and adaptive memory management, or by modifying loss functions and training strategies (e.g., fine-tuning sparse networks, logit masking). These advancements are crucial for building robust and adaptable AI systems capable of learning in dynamic real-world environments, with applications ranging from robotics and personalized medicine to continuously evolving data streams in various domains.