Research map / Probabilistic and scientific modeling / Variational and generative sampling
Tighter Variational Bounds are Not Necessarily Better
Tom Rainforth, Adam Kosiorek, Tuan Anh Le, Chris Maddison, Maximilian Igl, Frank Wood, Yee Whye Teh
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 MIWAE (8,8) instead of IWAE | MNIST training · MNIST test | IWAE-64: -86.11 ± 0.10 -> -85.69 ± 0.04 | Improved |
| Use PIWAE (8,8) instead of IWAE | MNIST training · MNIST test | log p̂(x): -84.52 ± 0.02 -> -84.46 ± 0.06 | Improved |
| Use MIWAE (8,8) instead of IWAE | MNIST training · MNIST test | -KL(Q||P): -1.59 ± 0.10 -> -0.72 ± 0.11 | Improved |
| Increase M in MIWAE while keeping total sample budget fixed (ablation) | MIWAE · MNIST test | log p̂(x) | Hurt |
| Increase M in MIWAE while keeping total sample budget fixed (ablation) | MIWAE · MNIST test | -KL(Q||P) | Improved |
| Use intermediate M and K rather than the endpoint settings (ablation) | MIWAE · MNIST test | IWAE-64 | Mixed |
| Increase M in PIWAE (ablation) | PIWAE · MNIST test | log p̂(x) | Improved |
| Increase β in CIWAE (ablation) | CIWAE · MNIST test | inference-network KL | Improved |
| Use K=1 and M=1000 rather than the IWAE configuration (ablation) | Gaussian toy model · Toy Gaussian optimization | optimization performance | Improved |
| Increase K (ablation) | IWAE inference network in the high-variance regime · High-variance Gaussian experiment | gradient SNR | Hurt |
Closest papers by meaning
- Debiasing Evidence Approximations: On Importance-weighted Autoencoders and Jackknife Variational Inference (ICLR 2018 · similarity 0.91)
- Inference Suboptimality in Variational Autoencoders (ICML 2018 · similarity 0.90)
- Iterative Amortized Inference (ICML 2018 · similarity 0.90)
- Hierarchical Importance Weighted Autoencoders (ICML 2019 · similarity 0.89)
- Amortized Inference Regularization (NeurIPS 2018 · similarity 0.89)
- Compute Optimal Inference and Provable Amortisation Gap in Sparse Autoencoders (ICML 2025 · 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.
