Research map / Label-efficient and robust learning / Test-time adaptation
Neural Priming for Sample-Efficient Adaptation
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 retrieval-based priming and attune the classifier | OpenCLIP ViT-B-16 · ImageNet zero-shot | top-1 accuracy: 2.45% | Helped |
| Add retrieval-based priming and attune the classifier | OpenCLIP ViT-B-16 · Transfer learning benchmarks, average across 6 datasets | top-1 accuracy: 4.25% | Helped |
| Add retrieval-based priming and attune the classifier | OpenCLIP ViT-B-16 · Few-shot transfer benchmarks | accuracy: 3.81% | Helped |
| Filter the priming pool using unlabeled test images | Neural Priming · ImageNet-V2, ImageNet-R, ImageNet Sketch, and ImageNet-A | accuracy: 1.09% | Helped |
| Use transductive priming instead of TPT | OpenCLIP ViT-B-16 · Four distribution-shift datasets | accuracy: 2.51% | Helped |
| Add Neural Priming to CuPL prompts | Zero-shot CLIP · FGVC Aircraft | accuracy: 3.78% | Helped |
| Increase the number of samples in the priming pool (ablation) | Neural Priming · Zero-shot downstream tasks | accuracy | Helped |
| Use a larger priming pool despite added retrieval noise (ablation) | Neural Priming · Rare classes in zero-shot tasks | accuracy | Mixed |
| Set sigma to 100 instead of 1 (ablation) | Neural Priming · ImageNet zero-shot | accuracy: 68.9 -> 70.75 | Helped |
| Set sigma to 100 instead of 1000 (ablation) | Neural Priming · FGVC Aircraft zero-shot | accuracy: 31.32 -> 33.03 | Helped |
| Set sigma to 100 instead of 1000 (ablation) | Neural Priming · Flowers102 zero-shot | accuracy: 78.81 -> 79.81 | Helped |
| Fine-tune only the last layer rather than all layers (ablation) | Neural Priming · Downstream tasks | performance | No clear effect |
Closest papers by meaning
- Quick-Tune: Quickly Learning Which Pretrained Model to Finetune and How (ICLR 2024 · similarity 0.85)
- Any-Shift Prompting for Generalization over Distributions (CVPR 2024 · similarity 0.84)
- Repeated Random Sampling for Minimizing the Time-to-Accuracy of Learning (ICLR 2024 · similarity 0.84)
- Robust Test-Time Adaptation for Zero-Shot Prompt Tuning (AAAI 2024 · similarity 0.84)
- On improved Conditioning Mechanisms and Pre-training Strategies for Diffusion Models (NeurIPS 2024 · similarity 0.84)
- A Closer Look at the Few-Shot Adaptation of Large Vision-Language Models (CVPR 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.