Research map / Probabilistic and scientific modeling / Variational and generative sampling
Hierarchical Importance Weighted Autoencoders
Chin-Wei Huang, Kris Sankaran, Eeshan Dhekane, Alexandre Lacoste, Aaron Courville
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| Use H-IWAE with α=1 instead of IWAE at K=10 | IWAE · Binarized MNIST test | NLL_te: 81.13 -> 81.04 | Improved |
| Use H-IWAE with α=0 instead of IWAE at K=10 | IWAE · Binarized OMNIGLOT test | NLL_te: 98.48 -> 98.80 | Hurt |
| Use H-IWAE with encoder updated twice per decoder update (ablation) | H-IWAE (α=1, K=5) · Caltech101 Silhouettes test | NLL_te: 106.62:0.09 -> 104.16:0.64 | Improved |
| Increase K from 1 to 10 for H-IWAE with α=1 (ablation) | H-IWAE · Binarized MNIST test | NLL_te: 81.89 -> 81.04 | Improved |
| Increase K from 1 to 10 for H-IWAE with α=1 (ablation) | H-IWAE · Binarized OMNIGLOT test | NLL_te: 100.57 -> 99.04 | Improved |
| Use arithmetic averaging (α=0) instead of α=1 at K=5 (ablation) | H-IWAE · Binarized MNIST test | NLL_te: 81.16 -> 81.93 | Hurt |
| Use balanced weighting (α=1) instead of arithmetic averaging (α=0) at K=5 (ablation) | H-IWAE · Binarized OMNIGLOT test | NLL_te: 99.23 -> 99.21 | No clear effect |
| Train with a shared common random number rather than independent latent factors (ablation) | Hierarchical importance sampler · Toy posterior experiment | variance of importance estimator | Improved |
| Use a Markov transition proposal instead of the hierarchical common-factor proposal (ablation) | Hierarchical importance sampler · Toy posterior experiment | proposal compensation | Hurt |
Closest papers by meaning
- Importance Weighting and Variational Inference (NeurIPS 2018 · similarity 0.89)
- Importance Weighted Hierarchical Variational Inference (NeurIPS 2019 · similarity 0.89)
- Tighter Variational Bounds are Not Necessarily Better (ICML 2018 · similarity 0.89)
- Variational Learning of Fractional Posteriors (ICML 2025 · similarity 0.87)
- Forward $\chi^2$ Divergence Based Variational Importance Sampling (ICLR 2024 · similarity 0.87)
- Beyond Vanilla Variational Autoencoders: Detecting Posterior Collapse in Conditional and Hierarchical Variational Autoencoders (ICLR 2024 · similarity 0.86)
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.
