Semantic Shift Estimation via Dual-Projection and Classifier Reconstruction for Exemplar-Free Class-Incremental Learning
TL;DR: This paper addresses semantic shift and decision bias in exemplar-free class-incremental learning via a dual-projection shift estimation and a classifier reconstruction.
Abstract: Exemplar-Free Class-Incremental Learning (EFCIL) aims to sequentially learn from distinct categories without retaining exemplars but easily suffers from catastrophic forgetting of learned knowledge. While existing EFCIL methods leverage knowledge distillation to alleviate forgetting, they still face two critical challenges: semantic shift and decision bias. Specifically, the embeddings of old tasks shift in the embedding space after learning new tasks, and the classifier becomes biased towards new tasks due to training solely with new data, hindering the balance between old and new knowledge. To address these issues, we propose the Dual-Projection Shift Estimation and Classifier Reconstruction (DPCR) approach for EFCIL. DPCR effectively estimates semantic shift through a dual-projection, which combines a learnable transformation with a row-space projection to capture both task-wise and category-wise shifts. Furthermore, to mitigate decision bias, DPCR employs ridge regression to reformulate a classifier reconstruction process. This reconstruction exploits previous in covariance and prototype of each class after calibration with estimated shift, thereby reducing decision bias. Extensive experiments demonstrate that, on various datasets, DPCR effectively balances old and new tasks, outperforming state-of-the-art EFCIL methods. Our codes are available at https://github.com/RHe502/ICML25-DPCR.
Lay Summary: We want to teach the computer to recognize new types of objects continually. However, the learned concepts are quickly forgotten when learning new objects. We attribute this to concept shift and model’s tendency to favor new task after learning new tasks. To overcome this problem, we learn a model combining two projections (TSSP and CIP) to rectify the concept shift during the continual learning. Also, the favor of new task is also corrected by a reconstruction process. These two components form our method DPCR and it achieves significant improvement over existing methods. Our proposed method can further alleviate the forgetting issue of models and obtain better models.
Link To Code: https://github.com/RHe502/ICML25-DPCR
Primary Area: Deep Learning
Keywords: Continual learning, Class-incremental learning, Exemplar-free, Semantic shift estimation
Submission Number: 2601
Loading