Reproduce “On the Discrimination and Consistency for Exemplar-Free Class Incremental Learning” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Tianqi Wang, Jingcai Guo, Depeng Li, Zhi Chen
DCNet introduces an exemplar-free class incremental learning method that constructs a discriminative and consistent feature space without relying on data replay. It achieves state-of-the-art performance on EF-CIL benchmarks, outperforming previous exemplar-free methods by up to 11.68% in accuracy and demonstrating competitive results against exemplar-based approaches.
Key insights
- Preserving a discriminative and consistent feature space is crucial for enabling effective inter-task interaction within the TIL+OOD framework in an exemplar-free setting.
- Theoretical analysis demonstrates that both increased inter-class separation and tighter intra-class aggregation are vital for robust Out-of-Distribution (OOD) detection and overall Class Incremental Learning performance.
- The dynamic adjustment of intra-class aggregation is necessary to counteract declining model plasticity in later tasks, ensuring consistent feature quality across incremental learning steps.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.