Research map / Probabilistic and scientific modeling / Variational and generative sampling
Lagging Inference Networks and Posterior Collapse in Variational Autoencoders
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 |
|---|---|---|---|
| Aggressively update the inference network before generator updates | VAE · Yahoo | NLL: 329.0 (0.1) -> 328.2 (0.2) | Helped |
| Aggressively update the inference network before generator updates | VAE · Yelp | NLL: 358.3 (0.2) -> 356.9 (0.2) | Helped |
| Aggressively update the inference network before generator updates | VAE · OMNIGLOT | NLL: 89.41 (0.04) -> 89.05 (0.05) | Helped |
| Add KL cost annealing to the aggressive inference-update method (ablation) | Ours · Yahoo | NLL: 328.2 (0.2) -> 326.7 (0.1) | Helped |
| Add KL cost annealing to the aggressive inference-update method (ablation) | Ours · Yelp | NLL: 356.9 (0.2) -> 355.9 (0.1) | Helped |
| Add KL cost annealing to the aggressive inference-update method (ablation) | Ours · OMNIGLOT | NLL: 89.05 (0.05) -> 89.11 (0.04) | Hurt |
| Limit inner inference updates to 10 iterations (ablation) | Ours · Yelp | NLL: 357.0 -> 357.9 | Hurt |
| Limit inner inference updates to 30 iterations (ablation) | Ours · Yelp | NLL: 357.0 -> 357.1 | No clear effect |
| Limit inner inference updates to 50 iterations (ablation) | Ours · Yelp | NLL: 357.0 -> 356.9 | No clear effect |
| Limit inner inference updates to 70 iterations (ablation) | Ours · Yelp | NLL: 357.0 -> 357.1 | No clear effect |
| Use a 100k-iteration KL annealing schedule instead of the original schedule (ablation) | VAE · Yahoo | NLL: 328.6 -> 327.5 | Helped |
| Use a 120k-iteration KL annealing schedule instead of the original schedule (ablation) | VAE · Yelp | NLL: 358.0 -> 356.9 | Helped |
Closest papers by meaning
- Beyond Vanilla Variational Autoencoders: Detecting Posterior Collapse in Conditional and Hierarchical Variational Autoencoders (ICLR 2024 · similarity 0.94)
- Don't Blame the ELBO! A Linear VAE Perspective on Posterior Collapse (NeurIPS 2019 · similarity 0.93)
- The Usual Suspects? Reassessing Blame for VAE Posterior Collapse (ICML 2020 · similarity 0.93)
- Posterior Collapse and Latent Variable Non-identifiability (NeurIPS 2021 · similarity 0.92)
- Gaussian Process Modeling of Approximate Inference Errors for Variational Autoencoders (CVPR 2022 · similarity 0.92)
- Preventing Posterior Collapse with delta-VAEs (ICLR 2019 · 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.