Research map / Label-efficient and robust learning / Recognition and long-tail
Network Dissection: Quantifying Interpretability of Deep Visual Representations
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Apply a random orthogonal change of basis | AlexNet Places205 conv5 · Broden concepts | unique detectors: 80% fewer | Hurt |
| Remove dropout in the FC layers (ablation) | Places205-AlexNet · Broden concepts, conv5 | texture detectors | Helped |
| Remove dropout in the FC layers (ablation) | Places205-AlexNet · Broden concepts, conv5 | object detectors | Hurt |
| Apply batch normalization at each convolutional layer (ablation) | Places205-AlexNet · Broden concepts | interpretability | Hurt |
| Increase conv5 width from 256 to 768 units and add global average pooling (ablation) | AlexNet-GAP-Wide · Places365 | unique detectors | Helped |
| Increase conv5 width from 256 to 768 units and add global average pooling (ablation) | AlexNet-GAP-Wide · Places365 | top-1 accuracy: ~0.5% top1 accuracy lower | No clear effect |
| Increase conv5 width further to 1024 or 2048 units (ablation) | AlexNet-GAP-Wide · Places365 | unique concepts | No clear effect |
| Train with a different random initialization for the same number of iterations (ablation) | Places205-AlexNet · Broden concepts | unique and total detector counts | No clear effect |
| Train for more iterations (ablation) | Places205-AlexNet · Broden concepts, conv5 | interpretability | Helped |
| Train AlexNet on Places rather than ImageNet | AlexNet conv5 · Broden concepts | unique detectors | Helped |
| Use deeper network architectures | CNN representations · Broden concepts, last convolutional layer | unique detectors | Helped |
| Use self-supervised rather than large-data supervised training | AlexNet-based CNNs · Broden concepts | object detectors | Hurt |
Closest papers by meaning
- PICNN: A Pathway towards Interpretable Convolutional Neural Networks (AAAI 2024 · similarity 0.90)
- Measuring Per-Unit Interpretability at Scale Without Humans (NeurIPS 2024 · similarity 0.90)
- Representational Similarity via Interpretable Visual Concepts (ICLR 2025 · similarity 0.89)
- A Convolutional Neural Network Interpretable Framework for Human Ventral Visual Pathway Representation (AAAI 2024 · similarity 0.89)
- Enhancing Pre-trained Representation Classifiability can Boost its Interpretability (ICLR 2025 · similarity 0.88)
- Visual Concept Connectome (VCC): Open World Concept Discovery and their Interlayer Connections in Deep Models (CVPR 2024 · similarity 0.88)
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.