Research map / Label-efficient and robust learning / Self-supervised and contrastive
Trading robust representations for sample complexity through self-supervised visual experience
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 |
|---|---|---|---|
| add orbit encoder rectification loss to orbit triplet loss (OJ vs OT) (ablation) | Orbit Joint loss · MNIST one-shot | accuracy: 0.37±0.04 -> 0.67±0.05 | Helped |
| add orbit triplet loss to orbit encoder loss (OJ vs OE) (ablation) | Orbit Joint loss · MNIST one-shot | accuracy: 0.40±0.05 -> 0.67±0.05 | Helped |
| orbit-based OJ instead of exemplar loss | embedding loss · MNIST one-shot | accuracy: 0.44±0.06 -> 0.67±0.05 | Helped |
| add spatial transformer modules with full or orbit supervision | ST and OT · MNIST one-shot | accuracy: 0.97±0.02 -> 0.97±0.01 | No clear effect |
| orbit-based OJ weak supervision instead of class-supervised triplet | embedding loss · Even/Odd MNIST transfer | accuracy: 0.49±0.08 -> 0.59±0.08 | Helped |
| OJ instead of OT or exemplar loss | embedding loss · NORB one-shot | accuracy: 0.55±0.12 -> 0.59±0.12 | No clear effect |
| OJ instead of OT | embedding loss · Multi-PIE verification | AUC: 0.92±0.01 -> 0.95±0.01 | Helped |
| OJ instead of OT | embedding loss · Multi-PIE retrieval | top-1 precision: 0.93±0.02 -> 0.91±0.01 | Hurt |
| OJ instead of exemplar loss | embedding loss · Multi-PIE retrieval | top-1 precision: 0.71±0.04 -> 0.91±0.01 | Helped |
| OJ embeddings learned from unlabeled video orbits | embedding loss · Late Night faces, 28-way linear SVM | accuracy | Helped |
| increase orbits for embedding learning from 10 to 500 and labeled examples 1 to 20 | OJ · Late Night faces | accuracy | Helped |
Closest papers by meaning
- Self-Supervised Learning from Structural Invariance (ICLR 2026 · similarity 0.88)
- Revisiting Self-Supervised Visual Representation Learning (CVPR 2019 · similarity 0.88)
- Self-supervised Learning is More Robust to Dataset Imbalance (ICLR 2022 · similarity 0.87)
- Revisiting Weakly Supervised Pre-Training of Visual Perception Models (CVPR 2022 · similarity 0.87)
- Self-supervised video pretraining yields robust and more human-aligned visual representations (NeurIPS 2023 · similarity 0.87)
- Self-Supervised Learning as Discrete Communication (ICML 2026 · similarity 0.87)
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.