Research map / Label-efficient and robust learning / Self-supervised and contrastive
Self-Supervised Relational Reasoning for Representation Learning
Read the paperCheck 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 |
|---|---|---|---|
| Train with intra-reasoning and inter-reasoning using the relation module | Relational Reasoning · CIFAR-100 linear evaluation (ResNet-32) | accuracy: 46.17$\pm$ vs 42.13$\pm$ | Helped |
| Train with intra-reasoning and inter-reasoning using the relation module | Relational Reasoning · tiny-ImageNet linear evaluation (ResNet-32) | accuracy: 30.54$\pm$ vs 25.79$\pm$ | Helped |
| Train with intra-reasoning and inter-reasoning using the relation module | Relational Reasoning · CIFAR-10 linear evaluation (ResNet-32) | accuracy: 74.99$\pm$ vs 77.02$\pm$ | Hurt |
| Increase the number of augmentations from K=2 to K=32 (ablation) | Relational Reasoning · CIFAR-10 linear evaluation (Conv-4) | accuracy: 55.31$\pm$ -> 60.33$\pm$ | Helped |
| Increase the number of augmentations from K=2 to K=32 (ablation) | Supervised baseline · CIFAR-10 linear evaluation (Conv-4) | accuracy: 79.61$\pm$ -> 80.00$\pm$ | No clear effect |
| Increase the number of augmentations from K=2 to K=32 (ablation) | RotationNet · CIFAR-10 linear evaluation (Conv-4) | accuracy: 51.58$\pm$ -> 52.25$\pm$ | No clear effect |
| Use concatenation instead of sum, mean, or maximum aggregation (ablation) | Relational Reasoning · Representation learning; benchmark details not provided | accuracy | Helped |
| Replace concatenation aggregation with maximum aggregation (ablation) | Relational Reasoning · Representation learning; benchmark details not provided | accuracy | Hurt |
| Replace the relation module with a dot product directly on representation pairs (ablation) | Relational Reasoning · Representation learning; benchmark details not provided | accuracy: -3% to -23% | Hurt |
| Replace the relation module with a second encoder stage and dot product (ablation) | Relational Reasoning · Representation learning; benchmark details not provided | accuracy: -3% to -23% | Hurt |
Closest papers by meaning
- Exploring Self-Distillation Based Relational Reasoning Training for Document-Level Relation Extraction (AAAI 2023 · similarity 0.88)
- Better Set Representations For Relational Reasoning (NeurIPS 2020 · similarity 0.88)
- Clustering Properties of Self-Supervised Learning (ICML 2025 · similarity 0.85)
- When No Paths Lead to Rome: Benchmarking Systematic Neural Relational Reasoning (NeurIPS 2025 · similarity 0.84)
- Systematic Relational Reasoning With Epistemic Graph Neural Networks (ICLR 2025 · similarity 0.84)
- Collapse-Proof Non-Contrastive Self-Supervised Learning (ICML 2025 · similarity 0.84)
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.