Research map / Probabilistic and scientific modeling / Variational and generative sampling
Meta-Amortized Variational Inference and Learning
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 |
|---|---|---|---|
| Doubly-amortize inference over related mixture distributions instead of training on one mixture | VAE inference · Unseen mixtures of Gaussians | clustering error: 27.9\% -> 9.9\% | Helped |
| Increase the number of training mixtures from 10 to 20 | MetaVAE · Unseen mixtures of Gaussians | clustering error: 21.2\% -> 15.8\% | Helped |
| Increase the number of training mixtures from 20 to 50 | MetaVAE · Unseen mixtures of Gaussians | clustering error: 15.8\% -> 9.9\% | Helped |
| Meta-compile inference across simulators rather than train a VAE on one simulator | VAE · Unseen mechanics simulators | MSE | Helped |
| Amortize inference over image transformations to learn transferable representations | Neural Statistician and Variational HomoEncoder · Transformed MNIST and Small NORB | classification accuracy: 10 to 35\% accuracy | Helped |
| Train the MetaVAE over a transformation family | MetaVAE representations · MNIST and NORB transformation invariance | L2 distance: Rotation: 1.65 vs 4.44; scale: 2.16 vs 5.44; skew: 1.47 vs 4.89; elevation: 0.39 vs 1.16; azimuth: 0.44 vs 1.42; lightin | Helped |
| Amortize over Gaussian, log-normal, and exponential distributions together rather than only one family | MetaVAE · Unseen Weibull, Laplace, and Beta distributions | MSE | Helped |
| Amortize over multiple exponential-family distributions instead of only the matching family | MetaVAE · Gaussian, log-normal, and exponential parameter inference | MSE | Mixed |
Closest papers by meaning
- Meta-Learning with Shared Amortized Variational Inference (ICML 2020 · similarity 0.93)
- Inference Suboptimality in Variational Autoencoders (ICML 2018 · similarity 0.92)
- Amortized Inference Regularization (NeurIPS 2018 · similarity 0.91)
- InfoVAE: Balancing Learning and Inference in Variational Autoencoders (AAAI 2019 · similarity 0.91)
- Iterative Amortized Inference (ICML 2018 · similarity 0.91)
- Generalization Gap in Amortized Inference (NeurIPS 2022 · similarity 0.91)
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.