Research map / Probabilistic and scientific modeling / Variational and generative sampling
Iterative Amortized 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 |
|---|---|---|---|
| Replace a standard one-pass encoder with iterative inference | Latent Gaussian models · MNIST, single-level | negative log likelihood (nats): 84.14 \pm 0.02 -> 83.84 \pm 0.05 | Helped |
| Replace a standard one-pass encoder with iterative inference | Latent Gaussian models · MNIST, hierarchical | negative log likelihood (nats): 82.63 \pm 0.01 -> 82.457 \pm 0.001 | Helped |
| Replace a standard one-pass encoder with iterative inference | Latent Gaussian models · CIFAR-10, single-level | negative log likelihood (bits/input dim.): 5.823 \pm 0.001 -> 5.64 \pm 0.03 | Helped |
| Replace a standard one-pass encoder with iterative inference | Latent Gaussian models · CIFAR-10, hierarchical | negative log likelihood (bits/input dim.): 5.565 \pm 0.002 -> 5.456 \pm 0.005 | Helped |
| Replace a standard one-pass encoder with iterative inference | Latent Gaussian models · RCV1 | perplexity: 323 \pm 3 -> 285.0 \pm 0.1 | Helped |
| Replace a standard one-pass encoder with iterative inference | Latent Gaussian models · RCV1 | perplexity (≤ column): 377.4 \pm 0.5 -> 314 \pm 1 | Helped |
| Increase the number of inference iterations (ablation) | Iterative inference models · MNIST | ELBO | Helped |
| Encode errors rather than approximate posterior gradients (ablation) | Iterative inference models · MNIST | ELBO | Helped |
| Add the data input to the encoded gradients (ablation) | Iterative inference models · MNIST | ELBO | Mixed |
| Increase the number of approximate-posterior samples (ablation) | Iterative inference models · MNIST | ELBO: more than 1 nat | Helped |
Closest papers by meaning
- Amortized Inference Regularization (NeurIPS 2018 · similarity 0.93)
- Inference Suboptimality in Variational Autoencoders (ICML 2018 · similarity 0.92)
- Gaussian Process Modeling of Approximate Inference Errors for Variational Autoencoders (CVPR 2022 · similarity 0.91)
- InfoVAE: Balancing Learning and Inference in Variational Autoencoders (AAAI 2019 · similarity 0.91)
- Compute Optimal Inference and Provable Amortisation Gap in Sparse Autoencoders (ICML 2025 · similarity 0.88)
- Leveraging Self-Consistency for Data-Efficient Amortized Bayesian Inference (ICML 2024 · 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.