Research map / Label-efficient and robust learning / Test-time adaptation
Fast and Flexible Multi-Task Classification using Conditional Neural Adaptive Processes
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 feature-extractor adaptation with task-specific FiLM parameters (ablation) | CNAPS · META-DATASET tasks, especially datasets unlike ILSVRC | accuracy: On MNIST: 76.0 ± 0.8 -> 92.7 ± 0.4 | Helped |
| add autoregressive information from earlier adapted layers (ablation) | CNAPS · Aircraft | accuracy: 76.8 ± 0.8 -> 80.5 ± 0.6 | Helped |
| add autoregressive information from earlier adapted layers (ablation) | CNAPS · Textures | accuracy: 62.5 ± 0.7 -> 58.3 ± 0.7 | Hurt |
| use an adaptation network instead of gradient-based feature adaptation | CNAPS · held-out 5-way tasks | test-time processing time: 1274 and 7214 seconds | Helped |
| use an adaptation network instead of gradient-based feature adaptation | CNAPS · held-out tasks with fewer shots | accuracy | Helped |
| use gradient descent for feature adaptation rather than amortized adaptation | CNAPS comparison · held-out tasks with large numbers of shots | accuracy | Mixed |
| deploy the trained classifier to continual learning | CNAPS · Split CIFAR100 | single-head accuracy: 37.2 vs 34.0 | Helped |
| deploy the trained classifier to continual learning | CNAPS · Split MNIST | single-head accuracy: 80.9 vs 82.5 | No clear effect |
| use CNAPS rather than prototypical networks for active learning | active learning classifier · Flowers and held-out Omniglot languages | accuracy | Helped |
| use variation-ratio or predictive-entropy acquisition rather than random acquisition | CNAPS active learning · Flowers and held-out Omniglot languages | accuracy | Helped |
Closest papers by meaning
- Autoencoding Conditional Neural Processes for Representation Learning (ICML 2024 · similarity 0.85)
- Task-Conditioned Adaptation of Visual Features in Multi-Task Policy Learning (CVPR 2024 · similarity 0.85)
- TADFormer: Task-Adaptive Dynamic TransFormer for Efficient Multi-Task Learning (CVPR 2025 · similarity 0.85)
- PALM: Pushing Adaptive Learning Rate Mechanisms for Continual Test-Time Adaptation (AAAI 2025 · similarity 0.84)
- Task-Adaptive Prompted Transformer for Cross-Domain Few-Shot Learning (AAAI 2024 · similarity 0.84)
- Rethinking Few-Shot Adaptation of Vision-Language Models in Two Stages (CVPR 2025 · 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.