Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Choose Your Neuron: Incorporating Domain Knowledge through Neuron-Importance
Ramprasaath R. Selvaraju, Prithvijit Chattopadhyay, Mohamed Elhoseiny, Tilak Sharma, Dhruv Batra, Devi Parikh, Stefan Lee
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 |
|---|---|---|---|
| finetune VGG16 on seen classes | NIWT-Attributes · CUB GZSL | harmonic mean H: 26.7% -> 37.0% | Improved |
| finetune VGG16 on seen classes | NIWT-Attributes · AWA2 GZSL | harmonic mean H: 36.1% -> 48.1% | Improved |
| finetune ResNet101 on seen classes | NIWT-Attributes · CUB GZSL | harmonic mean H: 17.3% -> 27.7% | Improved |
| finetune ResNet101 on seen classes | NIWT-Attributes · AWA2 GZSL | harmonic mean H: 27.5% -> 40.5% | Improved |
| add L2 regularization toward mean seen weights (ablation) | NIWT-Attributes (VGG16) · AWA2 GZSL | unseen accuracy: 33.9% -> 41.3% | Improved |
| increase regularization strength beyond 1e-4 (ablation) | NIWT-Attributes (VGG16) · AWA2 GZSL | unseen accuracy | Hurt |
| add regularization (seen accuracy cost) (ablation) | NIWT-Attributes (VGG16) · AWA2 GZSL | seen accuracy: about 3% lower seen accuracy | Hurt |
| extract importance at conv5_3 instead of conv4_3 (ablation) | NIWT-Attributes (VGG16) · AWA2 GZSL | harmonic mean H: 39.3 -> 48.1 | Improved |
| extract importance at conv5_3 instead of fc7 (ablation) | NIWT-Attributes (VGG16) · AWA2 GZSL | harmonic mean H: 1 -> 48.1 | Improved |
| extract importance at conv5_3 instead of fc6 (ablation) | NIWT-Attributes (VGG16) · AWA2 GZSL | harmonic mean H: 40.2 -> 48.1 | Improved |
| use ImageNet images instead of random normal noise for weight optimization (ablation) | NIWT-Attributes (VGG16) · CUB GZSL | harmonic mean H: 30.2 -> 37.0 | Improved |
| use seen-class images instead of ImageNet images (ablation) | NIWT-Attributes (VGG16) · CUB GZSL | harmonic mean H: 37.0 -> 38.1 | Improved |
Closest papers by meaning
- Knowledge-Aware Neuron Interpretation for Scene Classification (AAAI 2024 · similarity 0.87)
- LINE: LLM-based Iterative Neuron Explanations for Vision Models (NeurIPS 2026 · similarity 0.86)
- Understanding Distributed Representations of Concepts in Deep Neural Networks without Supervision (AAAI 2024 · similarity 0.86)
- WWW: A Unified Framework for Explaining What Where and Why of Neural Networks by Interpretation of Neuron Concepts (CVPR 2024 · similarity 0.85)
- NeuronEye: Query-Guided Visual Concept Activation for Vision-Language Reasoning (NeurIPS 2026 · similarity 0.85)
- Automated Natural Language Explanation of Deep Visual Neurons with Large Models (Student Abstract) (AAAI 2024 · 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.
