Research map / Probabilistic and scientific modeling / Variational and generative sampling
Understanding VAEs in Fisher-Shannon Plane
Huangjie Zheng, Jiangchao Yao, Ya Zhang, Ivor W. Tsang, Jia Wang
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 |
|---|---|---|---|
| Add Fisher-information regularization with F=20 | VAE, yielding FAE · MNIST | test NLL: 85.56 -> 79.30 | Improved |
| Set Fisher-information constraint to F=0 instead of F=20 (ablation) | FAE · MNIST | test NLL: 79.30 -> 83.24 | Hurt |
| Use Fisher-information constraint F=20 rather than F=0.1 (ablation) | FAE · Reversed MNIST | train NLL: 96.67 -> 94.48 | Improved |
| Use Fisher-information constraint F=20 rather than F=0.1 (ablation) | FAE · Reversed MNIST | test NLL: 108.37 -> 164.56 | Hurt |
| Lower the Fisher-information constraint from F=20 to F=0.1 (ablation) | FAE · Reversed MNIST | test NLL: 164.56 -> 108.37 | Improved |
| Increase the Fisher-information constraint (ablation) | FAE · MNIST and SVHN reconstruction | reconstruction detail and fidelity | Improved |
| Increase the Fisher-information constraint (ablation) | FAE · MNIST latent representation | disentanglement | Improved |
| Increase entropy power by reducing F (ablation) | FAE · Reversed MNIST | generalization: 164.56 -> 108.37 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- pixelVAE — PixelVAE: A Latent Variable Model for Natural Images (ICLR 2017) · better on 0 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Embrace the Gap: VAEs Perform Independent Mechanism Analysis (NeurIPS 2022 · similarity 0.87)
- Information-theoretic Generalization Analysis for VQ-VAEs: A Role of Latent Variables (NeurIPS 2025 · similarity 0.86)
- Poisson Variational Autoencoder (NeurIPS 2024 · similarity 0.85)
- A Statistical Analysis of Wasserstein Autoencoders for Intrinsically Low-dimensional Data (ICLR 2024 · similarity 0.84)
- Generalization in VAE and Diffusion Models: A Unified Information-Theoretic Analysis (ICLR 2025 · similarity 0.84)
- Be a Goldfish: Forgetting Bad Conditioning in Sparse Linear Regression via Variational Autoencoders (ICML 2025 · similarity 0.84)
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.
