Research map / Continual learning / Class-incremental
Decouple Your Discovery and Memory in Continual Generalized Category Discovery
Jiawei Yu, Zijian Gao, Xingxing Zhang, Xuan Liu, Huaimin Wang, Kele Xu
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 class-group entropic regularization (L_GER) (ablation) | discovery branch · CIFAR100 (6 stages) | new-class accuracy (%): 3.90 -> 42.06 | Improved |
| add class-group entropic regularization (L_GER) (ablation) | discovery branch · CUB (6 stages) | new-class accuracy (%): 4.81 -> 9.22 | Improved |
| add clustering-guided initialization (Init) (ablation) | discovery branch · CUB (6 stages) | new-class accuracy (%): 4.81 -> 35.33 | Improved |
| combine L_GER and Init (ablation) | discovery branch · CIFAR100 (6 stages) | All accuracy (%): 59.66 -> 64.00 | Improved |
| add recursive analytic memory branch (ablation) | DYDM without memory · CIFAR100 (6 stages) | All accuracy (%): 64.00 -> 78.96 | Improved |
| add knowledge rehearsal distillation (L_KD) from memory branch (ablation) | DYDM · CUB (6 stages) | new-class accuracy (%): 50.85 -> 56.61 | Improved |
| add knowledge rehearsal distillation (L_KD) from memory branch (ablation) | DYDM · CIFAR100 (6 stages) | All accuracy (%): 78.96 -> 79.72 | Improved |
| plug DYDM memory branch into Happy | Happy · CIFAR100 (6 stages) | Md novel discovery (%): 50.20 -> 68.74 | Improved |
| plug DYDM memory branch into SimGCD* | SimGCD* · CIFAR100 (6 stages) | F_base forgetting (%): 12.96 -> 2.52 | Improved |
| use feature distillation or Happy-style anti-forgetting versus finetuning | C-GCD baselines · CIFAR100 | plasticity | Hurt |
| increase expansion size of analytic classifier from 1k to 9k (ablation) | memory branch · CIFAR100 (6 stages) | All accuracy (%): 77.98 -> 80.11 | Improved |
| increase expansion size of analytic classifier from 1k to 9k (ablation) | memory branch · CUB200 (6 stages) | All accuracy (%): 73.49 -> 52.25 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- MetaGCD — MetaGCD: Learning to Continually Learn in Generalized Category Discovery (ICCV 2023) · better on 1 of 1 shared result
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
- Happy: A Debiased Learning Framework for Continual Generalized Category Discovery (NeurIPS 2024 · similarity 0.93)
- Continual Generalized Category Discovery: Learning and Forgetting from a Bayesian Perspective (ICML 2025 · similarity 0.93)
- Solving the Catastrophic Forgetting Problem in Generalized Category Discovery (CVPR 2024 · similarity 0.90)
- Adaptive Discovering and Merging for Incremental Novel Class Discovery (AAAI 2024 · similarity 0.89)
- Active Generalized Category Discovery (CVPR 2024 · similarity 0.88)
- When Domain Generalization meets Generalized Category Discovery: An Adaptive Task-Arithmetic Driven Approach (CVPR 2025 · similarity 0.88)
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.
