Research map / Probabilistic and scientific modeling / Variational and generative sampling
Amortized Inference Regularization
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 denoising or weight-normalization regularization to the inference model | VAE · MNIST, OMNIGLOT, and Caltech 101 Silhouettes | -ln pθ(x): 86.93 -> 86.42; 110.32 -> 109.16; 109.14 -> 108.64 | Helped |
| add denoising regularization with α=0.5 | IWAE · MNIST, OMNIGLOT, and Caltech 101 Silhouettes, k=8 | -ln pθ(x): 86.21 -> 85.78; 108.18 -> 107.01; 108.65 -> 107.34 | Helped |
| add denoising regularization with α=0.5 | IWAE · MNIST, OMNIGLOT, and Caltech 101 Silhouettes, k=64 | -ln pθ(x): 86.06 -> 85.55; 107.31 -> 106.02; 108.89 -> 106.94 | Helped |
| use a larger encoder (ablation) | IWAE · MNIST, k=1 | -ln pθ(x): 86.93 -> 87.43 | Hurt |
| use a larger encoder (ablation) | IWAE · MNIST, k=8 | -ln pθ(x): 86.21 -> 86.98 | Hurt |
| use a larger encoder (ablation) | IWAE · MNIST, k=64 | -ln pθ(x): 86.06 -> 86.70 | Hurt |
| increase AIR regularization strength too far (ablation) | VAE and IWAE · MNIST, OMNIGLOT, and Caltech 101 Silhouettes | test log-likelihood | Hurt |
| increase importance samples from k=1 to k=8 or k=64 while over-regularized (ablation) | AIR-regularized IWAE · MNIST, OMNIGLOT, and Caltech 101 Silhouettes | test log-likelihood | Helped |
Closest papers by meaning
- Generalization Gap in Amortized Inference (NeurIPS 2022 · similarity 0.94)
- Iterative Amortized Inference (ICML 2018 · similarity 0.93)
- Gaussian Process Modeling of Approximate Inference Errors for Variational Autoencoders (CVPR 2022 · similarity 0.93)
- Semi-Amortized Variational Autoencoders (ICML 2018 · similarity 0.92)
- InfoVAE: Balancing Learning and Inference in Variational Autoencoders (AAAI 2019 · similarity 0.92)
- Inference Suboptimality in Variational Autoencoders (ICML 2018 · similarity 0.92)
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.