Research map / Label-efficient and robust learning / Recognition and long-tail
Max-Margin Contrastive Learning
Anshul Shah, Suvrit Sra, Rama Chellappa, Anoop Cherian
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 |
|---|---|---|---|
| replace the contrastive loss with MMCL | ImageNet-100 linear evaluation · ImageNet-100 | top-1: 79.0 -> 80.7 | Improved |
| replace the contrastive loss with MMCL | ImageNet-1K linear evaluation · ImageNet-1K | top-1: 63.9 -> 63.8 | No clear effect |
| replace SimCLR loss with MMCL | STL-10 with batch size 128 · STL-10 | accuracy: 77.18 -> 86.8 | Improved |
| replace SimCLR loss with MMCL | CIFAR-100 with batch size 512 · CIFAR-100 | accuracy: 66.32 -> 70.7 | Improved |
| replace MoCo-v2 with MMCL | UCF-101 RGB · UCF-101 | top-1: 46.8 -> 52.45 | Improved |
| replace GraphCL loss with MMCL | MUTAG graph representation learning · MUTAG | accuracy: 86.80 ± 1.34 -> 88.42 ± 1.33 | Improved |
| replace the RBF kernel with a linear kernel (ablation) | MMCL · CIFAR-100 | readout accuracy: 55.35 % -> 41.43% | Hurt |
| replace the RBF kernel with a tanh kernel (ablation) | MMCL · CIFAR-100 | readout accuracy: 55.35 % -> 54.53% | Hurt |
| replace the RBF kernel with a linear kernel (ablation) | MMCL · STL-10 | readout accuracy: 81.33% -> 74.82% | Hurt |
| replace the RBF kernel with a tanh kernel (ablation) | MMCL · STL-10 | readout accuracy: 81.33% -> 80.5% | Hurt |
| use a small slack penalty C=0.1 (ablation) | MMCL trained on CIFAR-100 · CIFAR-100 | readout accuracy: 55.1 -> 44.41 | Hurt |
| use a small slack penalty C=0.1 (ablation) | MMCL trained on STL-10 · STL-10 | readout accuracy: 79.42 -> 73.82 | Hurt |
Closest papers by meaning
- Why do We Need Large Batchsizes in Contrastive Learning? A Gradient-Bias Perspective (NeurIPS 2022 · similarity 0.87)
- A Simple Framework for Contrastive Learning of Visual Representations (ICML 2020 · similarity 0.87)
- Bridging Mini-Batch and Asymptotic Analysis in Contrastive Learning: From InfoNCE to Kernel-Based Losses (ICML 2024 · similarity 0.87)
- BECLR: Batch Enhanced Contrastive Few-Shot Learning (ICLR 2024 · similarity 0.86)
- Linear Differential Vision Transformer: Learning Visual Contrasts via Pairwise Differentials (NeurIPS 2025 · similarity 0.86)
- Variational Supervised Contrastive Learning (NeurIPS 2025 · similarity 0.86)
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.
