Research map / Label-efficient and robust learning / Test-time adaptation
Learn to Categorize or Categorize to Learn? Self-Coding for Generalized Category Discovery
Sarah Rastegar, Hazel Doughty, Cees G. M. Snoek
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 code contrastive loss L_C_code (ablation) | InfoSieve (L_C_in baseline) · CUB-200 | accuracy (All): 66.8 -> 67.7 | Improved |
| add binary code constraint L_code_cond (ablation) | InfoSieve · CUB-200 | accuracy (All): 67.7 -> 68.5 | Improved |
| add code length loss L_length (ablation) | InfoSieve · CUB-200 | accuracy (Novel): 64.0 -> 64.5 novel; 77.5 -> 76.1 known | Mixed |
| add mask binary constraint L_mask_cond (ablation) | InfoSieve · CUB-200 | accuracy (Novel): 64.5 -> 62.3 novel; 76.1 -> 78.7 known | Mixed |
| add categorizer loss L_Cat to full set of other losses (ablation) | InfoSieve · CUB-200 | accuracy (All): 67.8 -> 69.4 | Improved |
| add L_Cat alone to contrastive input loss (ablation) | L_C_in baseline · CUB-200 | accuracy (Known): 78.1 -> 76.4 | Hurt |
| increase mask constraint coefficient mu from 0.01 to 2 (ablation) | InfoSieve · CUB-200 | accuracy (All): 69.4 -> 70.6 | Improved |
| vary code constraint coefficient zeta 0.01 to 2 (ablation) | InfoSieve · CUB-200 | accuracy (All): 69.4 -> 69.9 (zeta=1) | No clear effect |
| vary code length coefficient delta 0.01 to 2 (ablation) | InfoSieve · CUB-200 | accuracy (All): 69.4 -> 69.8 (delta=2) | No clear effect |
| vary code mapping coefficient eta 0.01 to 2 (ablation) | InfoSieve · CUB-200 | accuracy (All): 69.4 -> 68.4 (eta=0.1) | No clear effect |
| use InfoSieve instead of best prior method on fine-grained datasets | GCD benchmarks · CUB-200 | accuracy (All): 63.5 -> 69.4 | Improved |
| use InfoSieve instead of best prior method on coarse-grained datasets | GCD benchmarks · CIFAR-10 | accuracy (All): 97.9 -> 94.8 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- PromptCAL — PromptCAL: Contrastive Affinity Learning via Auxiliary Prompts for Generalized Novel Category Discovery (CVPR 2023) · better on 1 of 4 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
- Prior-Constrained Association Learning for Fine-Grained Generalized Category Discovery (AAAI 2025 · similarity 0.88)
- Flipped Classroom: Aligning Teacher Attention with Student in Generalized Category Discovery (NeurIPS 2024 · similarity 0.88)
- Unleashing the Potential of Model Bias for Generalized Category Discovery (AAAI 2025 · similarity 0.88)
- SEAL: Semantic-Aware Hierarchical Learning for Generalized Category Discovery (NeurIPS 2025 · similarity 0.87)
- Solving the Catastrophic Forgetting Problem in Generalized Category Discovery (CVPR 2024 · similarity 0.87)
- Prototypical Hash Encoding for On-the-Fly Fine-Grained Category Discovery (NeurIPS 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.
