Research map / Label-efficient and robust learning / Recognition and long-tail
Select, Hypothesize and Verify: Towards Verified Neuron Concept Interpretation
ZeBin Ji, Yang Hu, Xiuli Bi, Bo Liu, Bin Xiao
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 Verify (activation-rate filtering) (ablation) | SIEVE · ImageNet val, ResNet-50 | mean Activation Rate (%): 72.87 -> 86.29 | Improved |
| add Select (ratio-based sample selection) (ablation) | SIEVE · ImageNet val, ResNet-50 | mean Activation Rate (%): 77.90 -> 86.29 | Improved |
| add Cluster (agglomerative clustering of patches) (ablation) | SIEVE · ImageNet val, ResNet-50 | mean Activation Rate (%): 81.52 -> 86.29 | Improved |
| add Verify (ablation) | SIEVE · ImageNet val, ResNet-50 | CLIP cos: 0.7656 -> 0.7914 | Improved |
| add Select (ablation) | SIEVE · ImageNet val, ResNet-50 | CLIP cos: 0.7624 -> 0.7914 | Improved |
| add Cluster (ablation) | SIEVE · ImageNet val, ResNet-50 | CLIP cos: 0.7821 -> 0.7914 | Improved |
| add Select, Cluster and Verify to bare baseline (ablation) | SIEVE · ImageNet val, ResNet-50 | mean Activation Rate (%): 45.57 -> 86.29 | Improved |
| vary threshold beta from 4 to 12 (ablation) | SIEVE · ImageNet val, ResNet-50 | mean Activation Rate (%): 85.61 to 86.29 | No clear effect |
| beta=10 vs beta=8 (ablation) | SIEVE · ImageNet val, ResNet-50 | CLIP cos: 0.7805 -> 0.7914 | Mixed |
| replace Broden concept set with Common Words (ablation) | SIEVE · ImageNet val, ResNet-50 | mean Activation Rate (%): 85.73 -> 86.29 | No clear effect |
| SIEVE vs CLIP-Dissect under domain shift | SIEVE · Eurosat (domain shift) | mean Activation Rate (%): 43.16 -> 75.45 | Improved |
| domain shift from Places365 to Eurosat | SIEVE · ResNet-18 | mean Activation Rate (%): 84.40 -> 75.45 | Hurt |
Closest papers by meaning
- Evaluating Neuron Explanations: A Unified Framework with Sanity Checks (ICML 2025 · similarity 0.89)
- Linear Explanations for Individual Neurons (ICML 2024 · similarity 0.88)
- WWW: A Unified Framework for Explaining What Where and Why of Neural Networks by Interpretation of Neuron Concepts (CVPR 2024 · similarity 0.87)
- Neuron Activation Coverage: Rethinking Out-of-distribution Detection and Generalization (ICLR 2024 · similarity 0.86)
- NeurFlow: Interpreting Neural Networks through Neuron Groups and Functional Interactions (ICLR 2025 · similarity 0.85)
- Training Verification-Friendly Neural Networks via Neuron Behavior Consistency (AAAI 2025 · similarity 0.85)
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.
