Research map / Continual learning / Class-incremental
Novel Class Discovery without Forgetting
K J Joseph, Sujoy Paul, Gaurav Aggarwal, Soma Biswas, Piyush Rai, Kai Han, Vineeth N Balasubramanian
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 pseudo-latent replay (ablation) | Ours · CIFAR-100-20-80 | All accuracy: 28.86 -> 67.24 | Improved |
| add pseudo-latent replay (ablation) | Ours · CIFAR-10-5-5 | All accuracy: 62.6 -> 91.52 | Improved |
| add mutual information regularizer (ablation) | Ours · CIFAR-100-20-80 | Unlab accuracy: 43.34 -> 49.67 | Improved |
| add mutual information regularizer (ablation) | Ours · CIFAR-10-5-5 | All accuracy: 88.85 -> 91.52 | Improved |
| add mutual information regularizer (ablation) | UNO · CIFAR-100-80 | unlabeled clustering accuracy: 89.31 -> 91.26 | Improved |
| add mutual information regularizer (ablation) | UNO · CIFAR-100-50 | unlabeled clustering accuracy: 59.45 -> 61.23 | Improved |
| add mutual information regularizer (ablation) | UNO · CIFAR-10-5 | unlabeled clustering accuracy: 94.15 -> 94.43 | Improved |
| raise KCI threshold tau from 0.8 to 0.99 (ablation) | Ours · CIFAR-100-50-50 (generalized) | Lab / Unlab accuracy: Lab 47.85 -> 62.76, Unlab 51.95 -> 36.42 | Mixed |
| sample mixing coefficient from Beta(100,1) instead of Beta(1,100) (ablation) | Ours · CIFAR-100-80-20 | Unlab accuracy: 77.03 -> 12.76 | Hurt |
| sample mixing coefficient uniformly Beta(1,1) instead of Beta(1,100) (ablation) | Ours · CIFAR-100-20-80 | All accuracy: 67.24 -> 65.91 | No clear effect |
| increase number of labeled classes (20 to 80 of CIFAR-100) | Ours · CIFAR-100 (task-aware) | Unlab accuracy: 49.67 -> 77.03 | Improved |
| replace UNO with Ours | UNO · CIFAR-100-20-80 (task-aware) | All accuracy: 28.33 -> 67.24 | Improved |
Closest papers by meaning
- Adaptive Discovering and Merging for Incremental Novel Class Discovery (AAAI 2024 · similarity 0.92)
- Class-Incremental Novel Class Discovery (ECCV 2022 · similarity 0.90)
- Continual Generalized Category Discovery: Learning and Forgetting from a Bayesian Perspective (ICML 2025 · similarity 0.88)
- Solving the Catastrophic Forgetting Problem in Generalized Category Discovery (CVPR 2024 · similarity 0.87)
- Joint Class-level and Instance-level Relationship Modeling for Novel Class Discovery (AAAI 2025 · similarity 0.87)
- Happy: A Debiased Learning Framework for Continual Generalized Category Discovery (NeurIPS 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.
