Research map / Label-efficient and robust learning / Recognition and long-tail
Learning Interpretable Queries for Explainable Image Classification with Information Pursuit
Stefan Kolek, Aditya Chattopadhyay, Kwan Ho Ryan Chan, Hector Andrade-Loarca, Gitta Kutyniok, René Vidal
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 alternating instead of joint optimization (ablation) | K-Learned V-IP · CIFAR-10 (budget 10) | test accuracy: 87.00% -> 95.12% | Improved |
| use alternating instead of joint optimization (ablation) | K-Learned V-IP · CUB-200 (budget 100) | test accuracy: 72.14% -> 74.52% | Improved |
| use alternating instead of joint optimization (ablation) | K-Learned V-IP · Stanford-Cars (budget 100) | test accuracy: 79.18% -> 82.39% | Improved |
| use alternating instead of joint optimization (ablation) | K-Learned V-IP · CIFAR-100 (budget 50) | test accuracy: 75.67% -> 75.20% | No clear effect |
| use alternating instead of joint optimization (ablation) | K-Learned V-IP · ImageNet-100 (budget 50) | test accuracy: 83.88% -> 83.99% | No clear effect |
| add query answer quantization (hard binary answers) (ablation) | K-Learned V-IP · CIFAR-100 (budget 50) | test accuracy: 82.45% -> 75.20% | Hurt |
| add query answer quantization (hard binary answers) (ablation) | K-Learned V-IP · CUB-200 (budget 100) | test accuracy: 81.65% -> 74.52% | Hurt |
| add query answer quantization (hard binary answers) (ablation) | K-Learned V-IP · RIVAL-10 (budget 10) | test accuracy: 99.49% -> 98.73% | Hurt |
| add query quantization (nearest-neighbor projection) (ablation) | K-Learned V-IP · CIFAR-100 (budget 50) | test accuracy: 79.93% -> 75.20% | Hurt |
| add query quantization (nearest-neighbor projection) (ablation) | K-Learned V-IP · Stanford-Cars (budget 100) | test accuracy: 82.06% -> 82.39% | Hurt |
| add query quantization (nearest-neighbor projection) (ablation) | K-Learned V-IP · CUB-200 (budget 100) | test accuracy: 77.54% -> 74.52% | Hurt |
| vary dictionary initialization (K-LLM, K-Random, K-Medoids) (ablation) | K-Learned V-IP · six datasets | test accuracy: within 5 points of accuracy | No clear effect |
Closest papers by meaning
- Explainable Neural Networks with Guarantee: A Sparse Estimation Approach (AAAI 2025 · similarity 0.85)
- From Flat to Hierarchical: Extracting Sparse Representations with Matching Pursuit (NeurIPS 2025 · similarity 0.84)
- NeSyFOLD: A Framework for Interpretable Image Classification (AAAI 2024 · similarity 0.84)
- SLICE: Stabilized LIME for Consistent Explanations for Image Classification (CVPR 2024 · similarity 0.84)
- Conformal Information Pursuit for Interactively Guiding Large Language Models (NeurIPS 2025 · similarity 0.84)
- Interpreting CLIP with Sparse Linear Concept Embeddings (SpLiCE) (NeurIPS 2024 · similarity 0.83)
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.
