Research map / Label-efficient and robust learning / Recognition and long-tail
Neural Dependencies Emerging From Learning Massive Categories
Ruili Feng, Kecheng Zheng, Kai Zhu, Yujun Shen, Jian Zhao, Yukun Huang, Deli Zhao, Jingren Zhou, Michael Jordan, Zheng-Jun Zha
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 |
|---|---|---|---|
| Predict unseen categories with a learned linear combination of base-category logits | ResNet-50 · ImageNet-1k, 900→100 | accuracy: 68.47 -> 68.03 | Hurt |
| Predict unseen categories with a learned linear combination of base-category logits | ResNet-50 · ImageNet-1k, 950→50 | accuracy: 68.47 -> 68.45 | No clear effect |
| Predict unseen categories with a learned linear combination of base-category logits | ResNet-50 · ImageNet-1k, 999→1 | accuracy: 68.47 -> 68.46 | No clear effect |
| Predict unseen categories with a learned linear combination of base-category logits | ResNet-50 · ImageNet-1k, 999→1 (pos&neg) | accuracy: 60.70 -> 61.50 | Improved |
| Predict unseen categories with a learned linear combination of base-category logits | Swin-T · ImageNet-1k, 900→100 | accuracy: 71.49 -> 71.486 | No clear effect |
| Predict unseen categories with a learned linear combination of base-category logits | Swin-T · ImageNet-1k, 950→50 | accuracy: 71.49 -> 71.578 | No clear effect |
| Predict unseen categories with a learned linear combination of base-category logits | Swin-T · ImageNet-1k, 999→1 | accuracy: 71.49 -> 71.56 | No clear effect |
| Predict unseen categories with a learned linear combination of base-category logits | Swin-T · ImageNet-1k, 999→1 (pos&neg) | accuracy: 76.20 -> 78.00 | Improved |
| Add a covariance regularizer to suppress a selected neural dependency | ResNet-18 · ImageNet-1k | accuracy: 69.83 -> 70.12 | Improved |
| Add a covariance regularizer to suppress a selected neural dependency | ResNet-50 · ImageNet-1k | accuracy: 76.37 -> 76.66 | Improved |
| Add a covariance regularizer to suppress a selected neural dependency | ViT-S · ImageNet-1k | accuracy: 80.67 -> 81.52 | Improved |
| Add a covariance regularizer to suppress a selected neural dependency | Swin-T · ImageNet-1k | accuracy: 82.16 -> 82.18 | No clear effect |
Closest papers by meaning
- Explainable Neural Networks with Guarantee: A Sparse Estimation Approach (AAAI 2025 · similarity 0.84)
- Graph Structure Inference with BAM: Neural Dependency Processing via Bilinear Attention (NeurIPS 2024 · similarity 0.83)
- How Deep Networks Learn Sparse and Hierarchical Data: the Sparse Random Hierarchy Model (ICML 2024 · similarity 0.83)
- From Flat to Hierarchical: Extracting Sparse Representations with Matching Pursuit (NeurIPS 2025 · similarity 0.83)
- Going Beyond Neural Network Feature Similarity: The Network Feature Complexity and Its Interpretation Using Category Theory (ICLR 2024 · similarity 0.83)
- Projecting Assumptions: The Duality Between Sparse Autoencoders and Concept Geometry (NeurIPS 2025 · 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.
