Research map / Probabilistic and scientific modeling / Variational and generative sampling
Meta-Learning with Shared Amortized Variational Inference
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 |
|---|---|---|---|
| Train the latent classifier model with shared amortized variational inference instead of direct Monte Carlo estimation | VERSA-style few-shot classifier on miniImageNet · miniImageNet 5-way 5-shot | accuracy: 68.0 $\pm$ 0.2 -> 69.8 $\pm$ 0.2 | Helped |
| Train with shared amortized variational inference instead of direct Monte Carlo estimation | VERSA-style few-shot classifier on miniImageNet · miniImageNet 5-way 1-shot | accuracy: 52.5 $\pm$ 0.3 -> 52.4 $\pm$ 0.3 | No clear effect |
| Share the prior and posterior inference network rather than use separate networks (ablation) | SAMOVAR-base · miniImageNet 5-way 5-shot | accuracy: 66.6 $\pm$ 0.2 -> 69.8 $\pm$ 0.2 | Helped |
| Share the prior and posterior inference network rather than use separate networks (ablation) | SAMOVAR-base · miniImageNet 5-way 1-shot | accuracy: 50.8 $\pm$ 0.3 -> 52.4 $\pm$ 0.3 | Helped |
| Add scaled cosine similarity (ablation) | SAMOVAR/TADAM architecture · miniImageNet 5-way 5-shot | accuracy: 75.3 $\pm$ 0.2 -> 76.9 $\pm$ 0.2 | Helped |
| Add auxiliary co-training (ablation) | SAMOVAR/TADAM architecture · miniImageNet 5-way 5-shot | accuracy: 75.3 $\pm$ 0.2 -> 76.4 $\pm$ 0.2 | Helped |
| Add task embedding network (ablation) | SAMOVAR/TADAM architecture · miniImageNet 5-way 5-shot | accuracy: 75.3 $\pm$ 0.2 -> 74.9 $\pm$ 0.2 | Hurt |
| Add scaled cosine similarity, auxiliary co-training, and task embedding network (ablation) | SAMOVAR/TADAM architecture · miniImageNet 5-way 1-shot | accuracy: 59.3 $\pm$ 0.3 -> 60.7 $\pm$ 0.3 | Helped |
| Use Monte Carlo training with few latent samples rather than variational inference | Synthetic-data inference model · Synthetic Gaussian tasks | posterior variance estimation | Hurt |
| Increase the number of classifier-weight samples at prediction time instead of using the distribution mean (ablation) | SAMOVAR-SC-AT-TEN · miniImageNet 5-way 5-shot | accuracy: +0.1% with 10K samples | Helped |
| Increase the number of classifier-weight samples at prediction time instead of using the distribution mean (ablation) | SAMOVAR-SC-AT-TEN · miniImageNet 5-way 1-shot | accuracy: +0.4% accuracy with 10K samples | Helped |
Closest papers by meaning
- Meta-Amortized Variational Inference and Learning (AAAI 2020 · similarity 0.93)
- Amortized Bayesian Meta-Learning (ICLR 2019 · similarity 0.91)
- Amortising Inference and Meta-Learning Priors in Neural Networks (ICLR 2026 · similarity 0.90)
- Variational Learning of Fractional Posteriors (ICML 2025 · similarity 0.88)
- Bayesian Online Natural Gradient (BONG) (NeurIPS 2024 · similarity 0.88)
- SoftCVI: Contrastive variational inference with self-generated soft labels (ICLR 2025 · 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.