Adaptive Uncertainty Calibration and Exemplar Management for Bridging Open Set Recognition and Incremental Learning
Year: 2026
Type: article
Abstract: A desirable open world recognition system requires cycling through three subtasks: First, open set recognition (OSR), i.e., classifying the known (classes seen during training) and rejecting the unknown (unseen classes)online; Second, manually annotate these unknown as novel known classes; Third, incremental learning (IL), i.e., continual learning these novel classes and retaining the memory of old classes. However, most existing methods tackle OSR and IL separately and present nontrivial challenges. Specifically, OSR methods often corrupt raw inputs, causing information loss, while IL methods suffer from insufficient memory utilization, resulting in resource waste. Therefore, we propose an adaptive uncertainty calibration and exemplar management (ACM) method that jointly addresses both challenges to bridge OSR and IL using only mutual information. ACM mitigates information loss and resource waste through adaptive memory allocation and soft label assignment. Experimental and visualization results show that ACM is feasible and state-of-the-art.
Authors: Fulin Gao, Weimin Zhong, Zhixing Cao, Xin Peng, Zhi Li
Institution: East China University of Science and Technology
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
Estimated APC paid: $2,645
Awards: G1701155033, G2362273132
OpenAlex ID: W7204127631
