Research map / Label-efficient and robust learning / Semi-supervised learning
USB: A Unified Semi-supervised Learning Benchmark for Classification
Yidong Wang, Hao Chen, Yue Fan, Wang SUN, Ran Tao, Wenxin Hou, Renjie Wang, Linyi Yang, Zhi Zhou, Lan-Zhe Guo, Heli Qi, Zhen Wu and 10 more
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 |
|---|---|---|---|
| Use pre-trained ViT instead of training ResNet from scratch | CV SSL methods · USB CV tasks | training iterations: 1,000k to 200k | Improved |
| Use pre-trained ViT instead of training from scratch | CV SSL methods · CV tasks | generalization performance | Improved |
| Use pre-trained ViT rather than ViT trained without pre-training (ablation) | CV SSL methods · CIFAR-100, 400 labels | test performance | Improved |
| Use pre-trained ViT rather than WRN-28-8 trained from scratch (ablation) | CV SSL methods · CIFAR-100, 400 labels | convergence | Improved |
| Use a diverse set of tasks across domains when evaluating SSL | SSL algorithm evaluation · CV, NLP, and Audio tasks | cross-domain ranking consistency | Mixed |
| Use unlabeled data through SSL rather than labeled-only supervised training | SSL algorithms · USB tasks | test performance | Mixed |
| Apply SSL with extremely scarce labeled data | SSL algorithms · USB tasks | test performance | Hurt |
| Use ReMixMatch rather than other evaluated SSL methods | CV SSL methods · USB CV tasks | mean error rate (%): 26.43 | Improved |
| Use AdaMatch rather than other evaluated SSL methods | Audio SSL methods · USB Audio tasks | mean error rate (%): 23.61 | Improved |
| Use SimMatch rather than other evaluated SSL methods | NLP SSL methods · USB NLP tasks | mean error rate (%): 28.34 | Improved |
| Use raw waveform rather than Mel spectrogram input | Audio SSL models · Audio tasks | generalization performance | Hurt |
| Use HuBERT rather than Wav2Vec 2.0 | Audio SSL methods · Audio tasks other than human-voice Keyword Spotting | robustness | Improved |
Closest papers by meaning
- Realistic Evaluation of Deep Semi-Supervised Learning Algorithms (NeurIPS 2018 · similarity 0.91)
- Towards Realistic Semi-Supervised Learning (ECCV 2022 · similarity 0.89)
- A Realistic Evaluation of Semi-Supervised Learning for Fine-Grained Classification (CVPR 2021 · similarity 0.88)
- A Closer Look at Benchmarking Self-supervised Pre-training with Image Classification (IJCV 2025 · similarity 0.88)
- Revisiting Semi-Supervised Learning in the Era of Foundation Models (NeurIPS 2025 · similarity 0.88)
- Erasing the Bias: Fine-Tuning Foundation Models for Semi-Supervised Learning (ICML 2024 · 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.
