Research map / Continual learning / Class-incremental
An Optimal Transport-driven Approach for Cultivating Latent Space in Online Incremental Learning
Quyen Tran, Hai Nguyen, Quan Dao, Hoang Phan, Linh Van, Khoat Than, Dinh Phung, Dimitris Metaxas, Trung Le
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use OTC instead of best baseline on CIFAR-10 at smallest memory | online class-incremental learning · CIFAR-10, M=0.2k | final average accuracy: 63.5 -> 64.8 (vs BiC+AC) | Improved |
| use OTC on CIFAR-10 M=1k where it ties BiC+AC within noise | online class-incremental learning · CIFAR-10, M=1k | final average accuracy: 75.8±3.1 -> 76.1±2.0 | No clear effect |
| use OTC on Tiny-ImageNet | online class-incremental learning · Tiny-ImageNet, M=2k | final average accuracy: 18.5 -> 19.5 (vs GSA) | Improved |
| use OTC versus CoPE for forgetting | online class-incremental learning · Tiny-ImageNet, M=10k | average forgetting: 8.2 -> 16.5 (CoPE vs OTC) | Hurt |
| use OTC versus best baseline for forgetting on CIFAR-10 | online class-incremental learning · CIFAR-10, M=1k | average forgetting: 10.0 -> 9.8 (vs GSA) | Improved |
| use multiple centroids (4) per class instead of 1 (ablation) | OTC · CIFAR-10, M=1000 | average accuracy: 71.6 -> 75.9 | Improved |
| increase centroids beyond optimum (4 to 8) (ablation) | OTC · CIFAR-10, M=1000 | average accuracy: 75.9 -> 74.5 | Hurt |
| too many centroids with small memory (>3 at M=200) (ablation) | OTC · CIFAR-10, M=0.2k | average accuracy | Hurt |
| select replay buffer samples by MMOT centroids instead of randomly (ablation) | OTC (4 centroids) · CIFAR-10, M=1000 | average accuracy: 73.4 -> 75.9 | Improved |
| centroid-based buffer selection with a single centroid (ablation) | OTC (1 centroid) · CIFAR-10, M=1000 | average accuracy: 68.7 -> 71.6 | Improved |
Closest papers by meaning
- Non-exemplar Online Class-Incremental Continual Learning via Dual-Prototype Self-Augment and Refinement (AAAI 2024 · similarity 0.87)
- Enhancing Online Continual Learning with Plug-and-Play State Space Model and Class-Conditional Mixture of Discretization (CVPR 2025 · similarity 0.87)
- Adapter Merging with Centroid Prototype Mapping for Scalable Class-Incremental Learning (CVPR 2025 · similarity 0.86)
- Forgetting, Ignorance or Myopia: Revisiting Key Challenges in Online Continual Learning (NeurIPS 2024 · similarity 0.85)
- Dynamic Expansion Diffusion Learning for Lifelong Generative Modelling (AAAI 2025 · similarity 0.84)
- Order-Robust Class Incremental Learning: Graph-Driven Dynamic Similarity Grouping (CVPR 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
