Research map / Label-efficient and robust learning / Recognition and long-tail
Superclass Learning With Representation Enhancement
Zeyu Gan, Suyun Zhao, Jinlong Kang, Liyuan Shang, Hong Chen, Cuiping Li
This paper tried a common change: Larger batch size
increase batch size from 64 to 512 · CIFAR100-7 · accuracy: 76.7 -> 77.9 Improved
What 785 accepted papers report for this change
- 469 improved
- 107 no clear effect
- 48 mixed
- 161 hurt
- Predicting the Emergence of Induction Heads in Language Model Pretraining ICML 2026 · increase batch size (fixed context 1024) · Natural data (CC100) · induction head emergence point (updates): rho=-1.0, p<0.001Improved
- When RL Meets Adaptive Speculative Training: A Unified Training-Serving System ICML 2026 · Increase batch size from 1 to 32 · MiniMax M2.1 FP8 · speedup: 1.57x -> 1.25xHurt
- Untied Ulysses: Memory-Efficient Context Parallelism via Headwise Chunking ICML 2026 · increase per-device batch size to 10 using freed memory · Llama 3-8B, 128K · throughput (TPS): 2464.10 (Ulysses) -> 2526.77 (UPipe)Improved
See this change in detailMatched by the wording of each paper's own result, read by AI from the full text. Papers report gains more often than failures, so this shows what gets published, not the chance a change works.
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 SCLRE instead of SimCLR | ResNet-50 · CIFAR100-3 | accuracy: 79.0 -> 80.1 | Improved |
| use SCLRE instead of SimCLR | ResNet-50 · CIFAR100-4 | accuracy: 80.6 -> 84.0 | Improved |
| use SCLRE instead of SimCLR | ResNet-50 · CIFAR100-7 | accuracy: 73.9 -> 78.1 | Improved |
| use SCLRE instead of the strongest reported baseline, SimCLR | ResNet-50 · mini-ImageNet | accuracy: 90.1 -> 89.3 | Hurt |
| use SCLRE instead of SupCon | ResNet-50 · FMoW | accuracy: 58.3 -> 64.8 | Improved |
| use SCLRE instead of the baseline | ResNet-50 · Adience | accuracy: 65.7 -> 68.7 | Improved |
| use SCLRE instead of SimCLR | ResNet-50 · VOC | accuracy: 80.0 -> 81.5 | Improved |
| add cross-instance attention (CIA) to the baseline (ablation) | SCLRE on CIFAR100-7 · CIFAR100-7 | accuracy: 68.9 -> 69.2 | Improved |
| add contrastive adjustment loss together with CIA (ablation) | SCLRE on CIFAR100-7 · CIFAR100-7 | accuracy: 69.2 -> 75.1 | Improved |
| add targeted adjustment loss to CIA and contrastive adjustment loss (ablation) | SCLRE on CIFAR100-7 · CIFAR100-7 | accuracy: 75.1 -> 77.9 | Improved |
| remove cross-entropy contribution by setting α=1 (ablation) | SCLRE on CIFAR100-7 · CIFAR100-7 | accuracy: 75.1 -> - | Hurt |
| increase batch size from 64 to 512 (ablation) | SCLRE on CIFAR100-7 · CIFAR100-7 | accuracy: 76.7 -> 77.9 | Improved |
Closest papers by meaning
- Classification Done Right for Vision-Language Pre-Training (NeurIPS 2024 · similarity 0.85)
- Learning and Evaluating Representations for Deep One-Class Classification (ICLR 2021 · similarity 0.84)
- Improving Barely Supervised Learning by Discriminating Unlabeled Samples with Super-Class (NeurIPS 2022 · similarity 0.83)
- SuperDisco: Super-Class Discovery Improves Visual Recognition for the Long-Tail (CVPR 2023 · similarity 0.83)
- Super-Class Guided Transformer for Zero-Shot Attribute Classification (AAAI 2025 · similarity 0.83)
- Improving Generalized Zero-Shot Learning by Exploring the Diverse Semantics from External Class Names (CVPR 2024 · similarity 0.81)
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.
