Research map / Label-efficient and robust learning / Recognition and long-tail
InfoNCE Induces Gaussian Distribution
Roy Betser, Eyal Gofer, Meir Yossef Levi, Guy Gilboa
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 |
|---|---|---|---|
| train with InfoNCE instead of cross-entropy supervision (ablation) | ResNet-18 · CIFAR-10 | CV of norms: 0.5 -> 0.09 | Improved |
| train with InfoNCE instead of cross-entropy supervision (ablation) | ResNet-18 · CIFAR-10 | AD average statistic: 3.3 -> 0.43 | Improved |
| train with InfoNCE instead of cross-entropy supervision (ablation) | ResNet-18 · CIFAR-10 | DP fraction of normal features: 3.9% -> 94.5% | Improved |
| train the linear encoder longer, from epoch 0 to epoch 100 (ablation) | linear InfoNCE encoder · Synthetic binary | CV of norms: 0.36 -> 0.09 | Improved |
| train the linear encoder longer, from epoch 0 to epoch 100 (ablation) | linear InfoNCE encoder · Synthetic binary | AD fraction of normal features: 30% -> 97% | Improved |
| train the linear encoder longer, from epoch 0 to epoch 100 (ablation) | linear InfoNCE encoder · Synthetic binary | DP fraction of normal features: 15% -> 98% | Improved |
| use a self-supervised pretrained model instead of supervised ResNet-34 (ablation) | pretrained encoders · MS-COCO | AD fraction of normal features: 0.0% -> 97.0% | Improved |
| use a self-supervised pretrained model instead of supervised DenseNet (ablation) | pretrained encoders · MS-COCO | AD fraction of normal features: 42.2% -> 97.0% | Improved |
| increase batch size and dimension (ablation) | linear InfoNCE encoder · Synthetic Laplace | CV of norms | Improved |
| increase batch size or dimension (ablation) | linear InfoNCE encoder · Synthetic data | uniformity | Improved |
| apply post hoc whitening to representations (ablation) | CLIP and DINO · pretrained embeddings | uniformity | Improved |
| train longer on CIFAR-10 (ablation) | two-layer MLP InfoNCE · CIFAR-10 | fraction of coordinates passing DP test | Improved |
Closest papers by meaning
- Learning multivariate Gaussians with imperfect advice (ICML 2025 · similarity 0.85)
- Inference via Interpolation: Contrastive Representations Provably Enable Planning and Inference (NeurIPS 2024 · similarity 0.85)
- Generalization Guarantees for Representation Learning via Data-Dependent Gaussian Mixture Priors (ICLR 2025 · similarity 0.85)
- On the Importance of Gaussianizing Representations (ICML 2025 · similarity 0.84)
- Learning to Embed Distributions via Maximum Kernel Entropy (NeurIPS 2024 · similarity 0.83)
- Partial Information Decomposition via Normalizing Flows in Latent Gaussian Distributions (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.
