Research map / Label-efficient and robust learning / Recognition and long-tail
Parametric Information Maximization for Generalized Category Discovery
Florent Chiaroni, Jose Dolz, Ziko Imtiaz Masud, Amar Mitiche, Ismail Ben Ayed
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 |
|---|---|---|---|
| compute marginal entropy over labeled plus unlabeled data instead of unlabeled only (ablation) | PIM · CUB (fine-grained) | ACC All: 58.3 -> 62.7 | Improved |
| add supervision constraint (CE on labeled samples) (ablation) | PIM · CUB (fine-grained) | ACC All: 56.6 -> 62.7 | Improved |
| add marginal entropy term to conditional entropy objective (ablation) | PIM · CUB (fine-grained) | ACC All: 38.6 -> 62.7 | Improved |
| minimize conditional entropy only (no marginal entropy, no constraint) (ablation) | PIM · CUB (fine-grained) | ACC All: 6.1 vs 62.7 | Hurt |
| learn lambda via bi-level optimization instead of fixed standard MI (ablation) | PIM · Herbarium19 (long-tailed) / CUB | ACC All | Improved |
| PIM versus standard InfoMax baselines TIM on fine-grained CUB | TIM · CUB (fine-grained) | ACC All: 53.4 -> 62.7 | Improved |
| PIM versus ORCA ViTB16 on simple CIFAR-10 | PIM · CIFAR10 | ACC All: 94.7 vs 97.1 | Hurt |
| use PIM instead of k-means for estimating number of classes | Max-ACC class-number search · fine-grained datasets mean error | Err: 18% -> 15% | Improved |
| PIM versus Max-ACC (GCD) on CIFAR-100 class-number estimation | Max-ACC · CIFAR100 K estimation | Err: 0% vs 5% | Hurt |
Closest papers by meaning
- Towards Understanding Parametric Generalized Category Discovery on Graphs (ICML 2025 · similarity 0.90)
- Prior-Constrained Association Learning for Fine-Grained Generalized Category Discovery (AAAI 2025 · similarity 0.89)
- Generalized Category Discovery via Reciprocal Learning and Class-Wise Distribution Regularization (ICML 2025 · similarity 0.89)
- Active Generalized Category Discovery (CVPR 2024 · similarity 0.88)
- DebGCD: Debiased Learning with Distribution Guidance for Generalized Category Discovery (ICLR 2025 · similarity 0.87)
- Consistent Supervised-Unsupervised Alignment for Generalized Category Discovery (NeurIPS 2025 · 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.
