Add to Chrome
Add to Firefox
Add to EdgeNew Insights on Relieving Task-Recency Bias for Online Class Incremental Learning
Feb 16, 2023
To imitate the ability of keeping learning of human, continual learning which can learn from a never-ending data stream has attracted more interests recently. In all settings, the online class incremental learning (CIL), where incoming samples from data stream can be used only once, is more challenging and can be encountered more frequently in real world. Actually, the CIL faces a stability-plasticity dilemma, where the stability means the ability to preserve old knowledge while the plasticity denotes the ability to incorporate new knowledge. Although replay-based methods have shown exceptional promise, most of them concentrate on the strategy for updating and retrieving memory to keep stability at the expense of plasticity. To strike a preferable trade-off between stability and plasticity, we propose a Adaptive Focus Shifting algorithm (AFS), which dynamically adjusts focus to ambiguous samples and non-target logits in model learning. Through a deep analysis of the task-recency bias caused by class imbalance, we propose a revised focal loss to mainly keep stability. By utilizing a new weight function, the revised focal loss can pay more attention to current ambiguous samples, which can provide more information of the classification boundary. To promote plasticity, we introduce a virtual knowledge distillation. By designing a virtual teacher, it assigns more attention to non-target classes, which can surmount overconfidence and encourage model to focus on inter-class information. Extensive experiments on three popular datasets for CIL have shown the effectiveness of AFS. The code will be available at \url{https://github.com/czjghost/AFS}.
