Research map / Label-efficient and robust learning / Self-supervised and contrastive
Bayesian Gated Non-Negative Contrastive Learning
Peng Cui, Jiahao Zhang, Lijie Hu
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 probabilistic gating (BayesNCL, STE) to NCL | NCL · ImageNet-100 | semantic consistency: 14.93 -> 36.14 | Improved |
| add probabilistic gating (BayesNCL, STE) to NCL | NCL · CIFAR-100 | semantic consistency: 9.91 -> 22.02 | Improved |
| add probabilistic gating (BayesNCL, STE) to NCL | NCL · CIFAR-10 | semantic consistency: 53.82 -> 56.50 | Improved |
| use learned Bayesian gating instead of deterministic Top-k masking | NCL+TopK · ImageNet-100 | semantic consistency: 14.27 -> 36.14 | Improved |
| add Top-k masking to NCL | NCL · CIFAR-10 | semantic consistency: 53.82 -> 51.81 | No clear effect |
| use hard gating instead of soft gating (ablation) | BayesNCL · CIFAR-100 | semantic consistency: 8.54 -> 22.02 | Improved |
| add gradient detachment for the gating head (ablation) | BayesNCL · ImageNet-100 | semantic consistency: 22.29 -> 36.14 | Improved |
| use a 2-layer gating head instead of 1-layer (ablation) | BayesNCL · CIFAR-100 | semantic consistency: 26.70 -> 22.02 | Hurt |
| use a 2-layer gating head instead of 3-layer (ablation) | BayesNCL · CIFAR-100 | semantic consistency: 17.88 -> 22.02 | Improved |
| use STE instead of Gumbel-Sigmoid gating | BayesNCL · ImageNet-100 | semantic consistency: 11.66 -> 36.14 | Improved |
| add Bayesian gating to NCL | NCL (ResNet-50) · ImageNet-100 (ResNet-50) | semantic consistency: 6.39 -> 13.40 | Improved |
| add Bayesian gating to NCL | NCL · ImageNet-100 | linear probe Acc@1: 69.63 -> 70.44 | Improved |
Closest papers by meaning
- A Bayesian Nonparametric Framework For Learning Disentangled Representations (ICLR 2026 · similarity 0.86)
- Variational Supervised Contrastive Learning (NeurIPS 2025 · similarity 0.84)
- Non-negative Contrastive Learning (ICLR 2024 · similarity 0.84)
- Enhancing Uncertainty Estimation and Interpretability with Bayesian Non-negative Decision Layer (ICLR 2025 · similarity 0.84)
- Self-Supervised Contrastive Learning is Approximately Supervised Contrastive Learning (NeurIPS 2025 · similarity 0.84)
- SoftCVI: Contrastive variational inference with self-generated soft labels (ICLR 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.
