Research map / Label-efficient and robust learning / Semi-supervised learning
Towards Discriminability and Diversity: Batch Nuclear-Norm Maximization Under Label Insufficient Situations
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 |
|---|---|---|---|
| maximize the nuclear norm of unlabeled batch predictions instead of minimizing entropy | CIFAR-100 semi-supervised learning with ResNet · CIFAR-100, 5000 labels | accuracy: 40.92±0.18 -> 41.59±0.27 | Helped |
| maximize the nuclear norm of unlabeled batch predictions instead of minimizing entropy | CIFAR-100 semi-supervised learning with ResNet · CIFAR-100, 10000 labels | accuracy: 50.36±0.20 -> 51.07±0.24 | Helped |
| add BNM to VAT | VAT · CIFAR-100, 5000 labels | accuracy: 56.63±0.18 -> 57.43±0.24 | Helped |
| add BNM to VAT rather than EntMin | VAT · CIFAR-100, 10000 labels | accuracy: 64.48±0.22 -> 64.61±0.15 | Helped |
| use BNM rather than EntMin | unsupervised domain adaptation · Office-31, average | accuracy: 83.8 -> 87.1 | Helped |
| add BNM to CDAN | CDAN · Office-31, average | accuracy: 87.5 -> 88.6 | Helped |
| use BNM rather than EntMin | unsupervised domain adaptation · Office-Home, average | accuracy: 64.5 -> 67.9 | Helped |
| add BNM to CDAN | CDAN · Office-Home, average | accuracy: 65.8 -> 69.4 | Helped |
| use BNM rather than EntMin | unsupervised open domain recognition · I2AwA, unknown categories | accuracy: 7.2 -> 39.7 | Helped |
| increase BNM loss weight from λ=1 to λ=2 (ablation) | BNM · I2AwA, average known/unknown | accuracy: 63.7 -> 64.0 | Helped |
| increase BNM loss weight from λ=2 to λ=4 (ablation) | BNM · I2AwA, average known/unknown | accuracy: 64.0 -> 63.0 | Hurt |
Closest papers by meaning
- $\epsilon$-Softmax: Approximating One-Hot Vectors for Mitigating Label Noise (NeurIPS 2024 · similarity 0.84)
- Nuclear Norm Regularization for Deep Learning (NeurIPS 2024 · similarity 0.84)
- Taming the Sigmoid Bottleneck: Provably Argmaxable Sparse Multi-Label Classification (AAAI 2024 · similarity 0.83)
- Implicit Bias of Spectral Descent and Muon on Multiclass Separable Data (NeurIPS 2025 · similarity 0.83)
- Balancing Feature Similarity and Label Variability for Optimal Size-Aware One-shot Subset Selection (ICML 2024 · similarity 0.83)
- Navigating Extremes: Dynamic Sparsity in Large Output Spaces (NeurIPS 2024 · similarity 0.82)
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.