Research map / Continual learning / Class-incremental
Heterogeneous Forgetting Compensation for Class-Incremental Learning
Jiahua Dong, Wenqi Liang, Yang Cong, Gan Sun
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 task-semantic aggregation (TSA) block to the baseline (ablation) | HFC baseline (DyTox-style losses) · CIFAR-100 (T=5, B=0%) | average accuracy: 81.6 -> 83.1 | Improved |
| add gradient-balanced forgetting compensation (GFC) loss on top of TSA (ablation) | HFC baseline + TSA · CIFAR-100 (T=5, B=0%) | average accuracy: 83.1 -> 84.0 | Improved |
| add gradient-balanced relation distillation (GRD) loss on top of TSA and GFC (ablation) | HFC · CIFAR-100 (T=5, B=0%) | average accuracy: 84.0 -> 86.4 | Improved |
| add TSA, GFC and GRD to reduce forgetting heterogeneity (ablation) | HFC · CIFAR-100 (T=5, B=0%) | forgetting heterogeneity (FH): 128.9 -> 103.6 | Improved |
| apply FC and RD plug-and-play losses to PODNet | PODNet · CIFAR-100 (T=10, B=0%, ResNet-32) | top-1 average accuracy: 54.5 -> 60.0 | Improved |
| apply FC and RD plug-and-play losses to SS-IL | SS-IL · CIFAR-100 (T=10, B=0%, ResNet-32) | top-1 average accuracy: 59.0 -> 64.9 | Improved |
| apply FC and RD plug-and-play losses to AFC | AFC · CIFAR-100 (T=10, B=0%, ResNet-32) | top-1 average accuracy: 57.1 -> 63.8 | Improved |
| apply FC and RD plug-and-play losses to FOSTER | FOSTER · CIFAR-100 (T=10, B=0%, ResNet-32) | top-1 average accuracy: 68.7 -> 69.9 | Improved |
| apply FC and RD plug-and-play losses to DyTox with ViT-Tiny | DyTox · CIFAR-100 (T=10, B=0%, ViT-Tiny) | top-1 average accuracy: 73.5 -> 74.0 | Improved |
| apply FC and RD losses to AFC in first task | AFC · CIFAR-100 (T=10, B=0%, ResNet-32) | top-1 accuracy at first task: 90.9 -> 88.7 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- FOSTER — FOSTER: Feature Boosting and Compression for Class-Incremental Learning (ECCV 2022) · better on 3 of 3 shared results
- SS-IL — SS-IL: Separated Softmax for Incremental Learning (ICCV 2021) · better on 2 of 2 shared results
- iCaRL — iCaRL: Incremental Classifier and Representation Learning (CVPR 2017) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Defying Imbalanced Forgetting in Class Incremental Learning (AAAI 2024 · similarity 0.89)
- Long-Tail Class Incremental Learning via Independent Sub-prototype Construction (CVPR 2024 · similarity 0.89)
- Make Domain Shift a Catastrophic Forgetting Alleviator in Class-Incremental Learning (AAAI 2025 · similarity 0.88)
- Gradient Reweighting: Towards Imbalanced Class-Incremental Learning (CVPR 2024 · similarity 0.88)
- Federated Class-Incremental Learning: A Hybrid Approach Using Latent Exemplars and Data-Free Techniques to Address Local and Global Forgetting (ICLR 2025 · similarity 0.87)
- Multi-layer Rehearsal Feature Augmentation for Class-Incremental Learning (ICML 2024 · similarity 0.87)
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.
