Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
MetaModulation: Learning Variational Feature Hierarchies for Few-Shot Learning with Fewer Tasks
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 learnable deterministic modulation of batch-normalization scale and shift between tasks (ablation) | ProtoNet · miniImagenet-S, ISIC, Dermnet-S | accuracy (%): 36.26 -> 42.44; 50.72 -> 56.25; 58.56 -> 63.13; 66.25 -> 74.23; 44.21 -> 49.46; 60.33 -> 66.12 | Helped |
| Add probabilistic latent modulation rather than deterministic modulation (ablation) | MTM · miniImagenet-S 1-shot and 5-shot | accuracy (%): 42.44 -> 43.21; 56.25 -> 57.26 | Helped |
| Modulate all layers hierarchically rather than using flat single-layer VTM (ablation) | VTM · miniImagenet-S, ISIC, Dermnet-S | accuracy (%): 42.05 -> 43.21; 55.82 -> 57.26; 64.04 -> 65.16; 72.59 -> 76.40; 49.19 -> 50.45; 64.62 -> 67.05 | Helped |
| Modulate only a selected single layer rather than all layers (ablation) | MTM · miniImagenet-S 1-shot | accuracy (%): 42.44 vs. 41.30; 41.32; 41.31; 39.47; 39.98 | Hurt |
| Modulate only a selected single layer rather than all layers (ablation) | MTM · miniImagenet-S 5-shot | accuracy (%): 56.25 vs. 54.21; 54.30; 54.13; 52.62; 53.32 | Hurt |
| Increase the weight on original-task constraints using λ (ablation) | HVTM · miniImagenet-S and ISIC, 1-shot and 5-shot | accuracy (%): miniImagenet-S: 43.21 at 0.01 vs. 38.01 at 100; ISIC: 65.13 at 0.01 vs. 59.12 at 100 | Hurt |
| Use MetaModulation instead of ATU | few-task meta-learning · miniImagenet-S 1-shot; ISIC 1-shot and 5-shot; Dermnet-S 1-shot and 5-shot; Tabular Murris 1-shot | accuracy (%): 42.60 -> 43.21; 62.84 -> 65.61; 74.50 -> 76.40; 48.33 -> 50.45; 65.16 -> 67.05; 82.03 -> 83.13 | Helped |
| Use MetaModulation instead of ATU | few-task meta-learning · Tabular Murris 5-shot | accuracy (%): 91.42 -> 91.23 | Hurt |
Closest papers by meaning
- Learning to Learn Kernels with Variational Random Features (ICML 2020 · similarity 0.89)
- Bayesian Meta-Learning for the Few-Shot Setting via Deep Kernels (NeurIPS 2020 · similarity 0.88)
- Submodular Meta-Learning (NeurIPS 2020 · similarity 0.88)
- Meta-Transfer Learning for Few-Shot Learning (CVPR 2019 · similarity 0.88)
- Data-Centric Meta-Learning for Robust Few-Shot Generalization (CVPR 2026 · similarity 0.88)
- Meta-Learning of Neural Architectures for Few-Shot Learning (CVPR 2020 · similarity 0.88)
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.