Research map / Continual learning / Class-incremental
Multi-Centroid Task Descriptor for Dynamic Class Incremental Inference
Tenghao Cai, Zhizhong Zhang, Xin Tan, Yanyun Qu, Guannan Jiang, Chengjie Wang, Yuan Xie
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 |
|---|---|---|---|
| Add the multi-centroid dynamic inference gate to DER | DER · CIFAR100-B0, 5/10/20/50 steps | average accuracy (%): 76.80±0.79 -> 78.15±0.58; 75.36±0.36 -> 77.40±0.94; 74.09±0.33 -> 76.20±1.18; 72.41±0.36 -> 75.81±1.38 | Improved |
| Add the multi-centroid dynamic inference gate to DER | DER · CIFAR100-B50, 2/5/10 steps | average accuracy (%): 74.61±0.52 -> 76.72±0.62; 73.21±0.78 -> 76.19±0.29; 72.81±0.88 -> 75.43±0.38 | Improved |
| Add the multi-centroid dynamic inference gate to DER | DER · ImageNet100-B0S10 | top-1 average accuracy (%): 77.18 -> 80.46 | Improved |
| Add the multi-centroid dynamic inference gate to DER | DER · ImageNet1000-B0S10 | top-1 average accuracy (%): 68.84 -> 70.80 | Improved |
| Add the multi-centroid dynamic inference gate to DER | DER · ImageNet100-B50S10 | top-1 average accuracy (%): 78.20 -> 79.83 | Improved |
| Add RandAugment to the dynamic inference network training (ablation) | Dynamic inference gate · CIFAR100-B0S10 | average accuracy (%): 75.38 -> 75.64 | Improved |
| Add knowledge distillation with data augmentation (ablation) | Dynamic inference gate · CIFAR100-B0S10 | average accuracy (%): 75.64 -> 76.19 | Improved |
| Add logit adjustment to the gate and data augmentation (ablation) | Dynamic inference gate · CIFAR100-B0S10 | average accuracy (%): 75.64 -> 76.45 | Improved |
| Add data augmentation, knowledge distillation, and logit adjustment (ablation) | Dynamic inference gate · CIFAR100-B0S10 | average accuracy (%): 75.38 -> 77.40 | Improved |
| Add supernet knowledge distillation (ablation) | DER · CIFAR100-B0S10 | average accuracy (%): 76.88 -> 76.75 | Hurt |
| Use the multi-centroid descriptor instead of CCCG's task classifier (ablation) | Task-ID inference on CIFAR100-B0S10 · CIFAR100-B0S10 | average accuracy (%): 68.52 -> 77.40 | Improved |
| Use the multi-centroid descriptor instead of CCCG's task classifier (ablation) | Task-ID inference on CIFAR100-B50S10 · CIFAR100-B50S10 | average accuracy (%): 55.62 -> 75.17 | Improved |
Closest papers by meaning
- Adapter Merging with Centroid Prototype Mapping for Scalable Class-Incremental Learning (CVPR 2025 · similarity 0.88)
- Class Incremental Learning via Likelihood Ratio Based Task Prediction (ICLR 2024 · similarity 0.88)
- Dynamic Integration of Task-Specific Adapters for Class Incremental Learning (CVPR 2025 · similarity 0.87)
- Task-Agnostic Guided Feature Expansion for Class-Incremental Learning (CVPR 2025 · similarity 0.87)
- Tripartite Weight-Space Ensemble for Few-Shot Class-Incremental Learning (CVPR 2025 · similarity 0.86)
- eTag: Class-Incremental Learning via Embedding Distillation and Task-Oriented Generation (AAAI 2024 · similarity 0.86)
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.
