Research map / Label-efficient and robust learning / Self-supervised and contrastive
Self-Supervised Learning with an Information Maximization Criterion
Serdar Ozsoy, Shadi Hamdan, Sercan O Arik, Deniz Yuret, Alper Tunga Erdogan
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 |
|---|---|---|---|
| CorInfoMax vs best baseline on linear evaluation | CorInfoMax · CIFAR-10 | top-1 accuracy (%): 92.94 -> 93.18 | Improved |
| CorInfoMax vs best baseline on linear evaluation | CorInfoMax · CIFAR-100 | top-1 accuracy (%): 70.90 -> 71.61 | Improved |
| CorInfoMax vs best baseline on linear evaluation | CorInfoMax · Tiny ImageNet (ResNet-50) | top-1 accuracy (%): 49.86 -> 54.86 | Improved |
| CorInfoMax vs best baseline on linear evaluation | CorInfoMax · ImageNet-100 (ResNet-50) | top-1 accuracy (%): 81.6 -> 82.64 | Improved |
| CorInfoMax vs best baseline on linear evaluation | CorInfoMax · ImageNet-1K (ResNet-50) | top-1 accuracy (%): 69.3 -> 69.08 | No clear effect |
| CorInfoMax vs VICReg semi-supervised, 10% labels | CorInfoMax · ImageNet-1K | top-1 accuracy (%): 62.16 -> 64.36 | Improved |
| CorInfoMax vs VICReg semi-supervised, 1% labels | CorInfoMax · ImageNet-1K | top-1 accuracy (%): 44.75 -> 44.89 | No clear effect |
| lower attraction coefficient alpha from 1000 to 125 (ablation) | CorInfoMax · CIFAR-100 (ResNet-18) | top-1 accuracy (%): 71.61 -> 69.04 | Hurt |
| raise attraction coefficient alpha from 1000 to 2000 (ablation) | CorInfoMax · CIFAR-100 (ResNet-18) | top-1 accuracy (%): 71.61 -> 69.96 | Hurt |
| alpha 250 or 500 instead of 1000 (ablation) | CorInfoMax · CIFAR-100 (ResNet-18) | top-1 accuracy (%): 71.61 -> 71.19 | No clear effect |
| batch size 256 instead of 512 (ablation) | CorInfoMax · CIFAR-100 (ResNet-18) | top-1 accuracy (%): 71.61 -> 70.32 | Hurt |
| batch size 1024 instead of 512 (ablation) | CorInfoMax · CIFAR-100 (ResNet-18) | top-1 accuracy (%): 71.61 -> 69.98 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- VICReg — VICReg: Variance-Invariance-Covariance Regularization for Self-Supervised Learning (ICLR 2022) · better on 6 of 6 shared results
- Barlow Twins — Barlow Twins: Self-Supervised Learning via Redundancy Reduction (ICML 2021) · better on 4 of 4 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- On Mutual Information Maximization for Representation Learning (ICLR 2020 · similarity 0.87)
- Preventing Dimensional Collapse in Self-Supervised Learning via Orthogonality Regularization (NeurIPS 2024 · similarity 0.87)
- Self-Supervised Learning via Maximum Entropy Coding (NeurIPS 2022 · similarity 0.87)
- The Role of Entropy and Reconstruction in Multi-View Self-Supervised Learning (ICML 2023 · similarity 0.86)
- Collapse-Proof Non-Contrastive Self-Supervised Learning (ICML 2025 · similarity 0.85)
- Unveiling the Dynamics of Information Interplay in Supervised Learning (ICML 2024 · similarity 0.85)
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.
