Research map / Label-efficient and robust learning / Self-supervised and contrastive
Self-supervised Label Augmentation via Input Transformations
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 |
|---|---|---|---|
| Replace invariant-label rotation training with joint original-and-transformation labels | CIFAR10 classification · CIFAR10 | accuracy: 90.79 -> 92.50 | Helped |
| Replace invariant-label rotation training with joint original-and-transformation labels | CIFAR100 classification · CIFAR100 | accuracy: 66.10 -> 68.68 | Helped |
| Add rotation-based SLA aggregation | CIFAR100 classification · CIFAR100 | accuracy: 68.27 -> 74.14 | Helped |
| Use color permutation instead of rotation for SLA aggregation | CIFAR10 classification · CIFAR10 | accuracy: 92.39 -> 92.51 | Hurt |
| Add SLA+SD to CutMix | PyramidNet200 · CIFAR10 | error rate: 2.88 -> 1.80 | Helped |
| Add SLA+SD to LDAM-DRW | Imbalanced CIFAR100, ratio 100 · CIFAR100 imbalanced | accuracy: 42.04 -> 45.53 | Helped |
| Add SLA+AG to ProtoNet | ProtoNet few-shot classifier · FC100 5-shot | accuracy: \ms{52.5}{0.6} -> \ms{55.7}{0.6} | Helped |
| Add SLA+AG to MetaOptNet-RR | MetaOptNet-RR few-shot classifier · FC100 5-shot | accuracy: \ms{55.3}{0.6} -> \ms{59.2}{0.5} | Helped |
| Add the full set of rotations and three color permutations (12 transformations) rather than rotation only (ablation) | SLA+AG on CUB200 · CUB200 | accuracy: 64.41 -> 65.53 | Helped |
| Increase composed transformations from 12 to 24 (ablation) | SLA+AG on Stanford Dogs · Stanford Dogs | accuracy: 69.95 -> 67.80 | Hurt |
| Use data augmentation with rotation while preserving the original labels (ablation) | CIFAR100 classification · CIFAR100 | accuracy: 68.27 -> 65.73 | Hurt |
| Add a separate rotation prediction task while retaining the primary labels (ablation) | CIFAR100 classification · CIFAR100 | accuracy: 68.27 -> 66.10 | Hurt |
Closest papers by meaning
- Boost Self-Supervised Dataset Distillation via Parameterization, Predefined Augmentation, and Approximation (ICLR 2025 · similarity 0.88)
- Self Iterative Label Refinement via Robust Unlabeled Learning (NeurIPS 2025 · similarity 0.87)
- Learning with Partial-Label and Unlabeled Data: A Uniform Treatment for Supervision Redundancy and Insufficiency (ICML 2024 · similarity 0.87)
- Distilling Reliable Knowledge for Instance-Dependent Partial Label Learning (AAAI 2024 · similarity 0.87)
- Rethinking Self-Distillation: Label Averaging and Enhanced Soft Label Refinement with Partial Labels (ICLR 2025 · similarity 0.86)
- Towards Understanding Self-Pretraining for Sequence Classification (NeurIPS 2026 · 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.