Research map / Label-efficient and robust learning / Recognition and long-tail
Discover-then-Name: Task-Agnostic Concept Bottlenecks via Automated Concept Discovery
Sukrut Rao, Sweta Mahajan, Moritz Böhle, Bernt Schiele
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 |
|---|---|---|---|
| use a single task-agnostic SAE concept layer instead of task-specific LLM-generated concept sets | DN-CBM · Places365 (ResNet-50) | accuracy (%): 52.7 -> 53.5 | Improved |
| task-agnostic concepts vs LaBo task-specific concepts on CIFAR100 | DN-CBM · CIFAR100 (ResNet-50) | accuracy (%): 69.1 vs 67.5 | Hurt |
| select a separate SAE per downstream dataset instead of a shared one (ablation) | DN-CBM · CIFAR100 (ResNet-50) | accuracy (%): 67.5 -> 69.2 | Improved |
| intervene to keep only bird concepts in the concept bottleneck (ablation) | DN-CBM · Waterbirds-100 | overall accuracy (%): 82.8 -> 89.4 | Improved |
| intervene to keep only bird concepts (ablation) | DN-CBM · Waterbirds-100 Landbird on Water | accuracy (%): 71.3 -> 86.6 | Improved |
| intervene to remove bird concepts (ablation) | DN-CBM · Waterbirds-100 | overall accuracy (%): 82.8 -> 60.8 | Hurt |
| use SAE nodes instead of CLIP neurons for concepts, named with same vocabulary (ablation) | DN-CBM · user study, semantic consistency | consistency and name accuracy ratings | Improved |
| raise the L1 sparsity coefficient of the SAE (ablation) | SAE · reconstruction error | reconstruction error | Mixed |
| unsupervised SAE concept naming vs CLIP retrieval on SUNAttributes | DN-CBM · SUNAttributes | Jaccard index: 22.0 vs 18.3 | Hurt |
Closest papers by meaning
- Concepts Worth Having: Refining VLM-Guided Concept Bottleneck Models with Minimal Annotations (NeurIPS 2026 · similarity 0.91)
- Coarse-to-Fine Concept Bottleneck Models (NeurIPS 2024 · similarity 0.91)
- Hybrid Concept Bottleneck Models (CVPR 2025 · similarity 0.91)
- Energy-Based Concept Bottleneck Models: Unifying Prediction, Concept Intervention, and Probabilistic Interpretations (ICLR 2024 · similarity 0.90)
- V2C-CBM: Building Concept Bottlenecks with Vision-to-Concept Tokenizer (AAAI 2025 · similarity 0.90)
- VLG-CBM: Training Concept Bottleneck Models with Vision-Language Guidance (NeurIPS 2024 · similarity 0.90)
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.
