Research map / Label-efficient and robust learning / Recognition and long-tail
A Simple Episodic Linear Probe Improves Visual Recognition in the Wild
Check 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 ELP-SR regularization | ResNet50 · CUB-200-2011 | accuracy: 85.5 -> 88.8 | Helped |
| Add ELP-SR regularization | ResNet50 · Stanford Cars | accuracy: 92.7 -> 94.2 | Helped |
| Add ELP-SR regularization | ResNet50 · FGVC-Aircraft | accuracy: 90.3 -> 92.7 | Helped |
| Add ELP-SR to ERM with best-setting τ-normalization | ResNet-32 ERM · CIFAR-10 long-tailed, imbalance ratio 100 | top-1 accuracy: 70.4 -> 78.0 | Helped |
| Add ELP-SR | ResNet-32 LDAM · CIFAR-100 long-tailed, imbalance ratio 100 | top-1 accuracy: 42.0 -> 43.9 | Helped |
| Add ELP-SR regularization | ResNet-50 · ImageNet-1K val | top-1 accuracy: 76.13 -> 76.82 | Helped |
| Use γ=2 rather than γ=3 (ablation) | ELP-SR · CUB-200-2011 | accuracy: 88.5 vs 88.8 | No clear effect |
| Replace the ELP-SR formulation with L1 or L2 classifier-probability distillation (ablation) | ResNet-50 · ImageNet-1K val | top-1 accuracy: 76.82 -> 76.12; 76.82 -> 76.18 | Hurt |
| Replace episodic ELP with a momentum-updated memory bank (ablation) | ResNet50 · CUB-200-2011 | accuracy: 86.1% and 86.5% | Hurt |
Closest papers by meaning
- LookWhere? Efficient Visual Recognition by Learning Where to Look and What to See from Self-Supervision (NeurIPS 2025 · similarity 0.85)
- Improving Visual Prompt Tuning by Gaussian Neighborhood Minimization for Long-Tailed Visual Recognition (NeurIPS 2024 · similarity 0.85)
- L-WISE: Boosting Human Visual Category Learning Through Model-Based Image Selection and Enhancement (ICLR 2025 · similarity 0.85)
- AutoVP: An Automated Visual Prompting Framework and Benchmark (ICLR 2024 · similarity 0.84)
- Closed-Form Linear-Probe Dataset Distillation for Pre-trained Vision Models (NeurIPS 2026 · similarity 0.84)
- ADDP: Learning General Representations for Image Recognition and Generation with Alternating Denoising Diffusion Process (ICLR 2024 · 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.