Research map / Probabilistic and scientific modeling / Variational and generative sampling
Iterative Refinement of the Approximate Posterior for Directed Belief Networks
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 |
|---|---|---|---|
| Refine posterior with AIR at test time (RWS+ vs RWS) | RWS-trained SBN 200 · MNIST | test log-likelihood estimate (nats): 102.51 -> 102.00 | Helped |
| Refine posterior with AIR at test time (RWS+ vs RWS) | RWS-trained SBN 200 · Caltech-101 Silhouettes | test log-likelihood estimate (nats): 121.38 -> 118.63 | Helped |
| Refine posterior with AIR at test time (RWS+ vs RWS) | RWS-trained SBN 200-200 · Caltech-101 Silhouettes | test log-likelihood estimate (nats): 112.86 -> 107.20 | Helped |
| Refine posterior with AIR at test time (RWS+ vs RWS) | RWS-trained SBN 200-200-200 · Caltech-101 Silhouettes | test log-likelihood estimate (nats): 110.57 -> 104.54 | Helped |
| Train with AIR instead of RWS | SBN 200 · MNIST | test log-likelihood estimate (nats): 102.51 -> 100.92 | Helped |
| Train with AIR instead of RWS | SBN 200-200-200 · MNIST | test log-likelihood estimate (nats): 92.00 -> 92.56 | Hurt |
| Train with AIR instead of RWS | DARN 500 · MNIST | test log-likelihood estimate (nats): 85.40 -> 85.46 | Hurt |
| Train with AIR instead of RWS | SBN 200-200 · Caltech-101 Silhouettes | test log-likelihood estimate (nats): 112.86 -> 106.94 | Helped |
| Use 20 instead of 5 AIR refinement steps in training (ablation) | AIR · DBN training, ESS and log-likelihood | test log-likelihood estimate (nats) | Helped |
| Add reweighting and inclusive KL (AIR+RW, AIR+RW+IKL) to AIR with 20 steps (ablation) | AIR(20) · final log-likelihood | test log-likelihood estimate (nats) | Hurt |
| Add reweighting to AIR with 5 steps (ablation) | AIR(5) · final log-likelihood | test log-likelihood estimate (nats) | Helped |
| Refine with 100 AIR steps vs no refinement | 1-layer SBN 20 units · exact log-probability -127.474 | test log-likelihood estimate (nats): -127.51 -> -127.48 | Helped |
Closest papers by meaning
- Inference Suboptimality in Variational Autoencoders (ICML 2018 · similarity 0.88)
- Undirected Graphical Models as Approximate Posteriors (ICML 2020 · similarity 0.87)
- VIKING: Deep variational inference with stochastic projections (NeurIPS 2025 · similarity 0.87)
- On the Expressiveness of Approximate Inference in Bayesian Neural Networks (NeurIPS 2020 · similarity 0.87)
- Do Bayesian Neural Networks Actually Behave Like Bayesian Models? (ICML 2025 · similarity 0.86)
- Variational Learning of Fractional Posteriors (ICML 2025 · similarity 0.85)
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.