Research map / Probabilistic and scientific modeling / Variational and generative sampling
Denoising Criterion for Variational Auto-Encoding Framework
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 |
|---|---|---|---|
| Add salt-and-pepper input corruption at noise level 5 | VAE · Binarized MNIST | negative variational lower bound: 96.14 ± 0.09 -> 95.52 ± 0.12* | Helped |
| Add salt-and-pepper input corruption at noise level 5 | IWAE · Binarized MNIST | negative variational lower bound: 94.36 ± 0.07 -> 93.67 ± 0.10* | Helped |
| Use noise level 15 instead of no input corruption (ablation) | VAE · Binarized MNIST | negative variational lower bound: 96.14 ± 0.09 -> 96.83 ± 0.17 | Hurt |
| Add Gaussian input corruption at noise level 7.5 | VAE · Frey Face | negative variational lower bound: 1304.79 ± 5.71 -> 1293.07 ± 5.03 | Helped |
| Add Gaussian input corruption at noise level 5 | IWAE · Frey Face | negative variational lower bound: 1318.04 ± 2.83 -> 1333.44 ± 2.74* | Hurt |
| Add salt-and-pepper input corruption at noise level 5 | VAE with GRU inference network · Binarized MNIST | negative variational lower bound: 96.07 ± 0.17 -> 94.30 ± 0.09* | Helped |
| Add salt-and-pepper input corruption at noise level 10 | IWAE with GRU inference network · Binarized MNIST | negative variational lower bound: 93.94 ± 0.06 -> 92.84 ± 0.07* | Helped |
| Increase the number of corruption samples M from 1 to 10 (ablation) | VAE with denoising criterion · MNIST | negative log-likelihood: 94.97, 94.44, and 94.39 | Helped |
| Increase the number of importance samples M from 1 to 10 (ablation) | IWAE with denoising criterion · MNIST | negative log-likelihood: 93.17, 92.89, and 92.85 | Helped |
| Replace the feedforward inference network with a GRU (ablation) | VAE · MNIST | negative variational lower bound | Hurt |
| Use data augmentation instead of denoising (ablation) | VAE · MNIST | negative log-likelihood: 94.32 ± 0.37 -> 93.88 ± 0.08 | Helped |
| Use data augmentation instead of denoising (ablation) | IWAE · MNIST | negative log-likelihood: 93.83 ± 0.06 -> 92.51 ± 0.07 | Hurt |
Closest papers by meaning
- Adversarial Symmetric Variational Autoencoder (NeurIPS 2017 · similarity 0.89)
- Explicitly Minimizing the Blur Error of Variational Autoencoders (ICLR 2023 · similarity 0.88)
- Denoising Masked Autoencoders Help Robust Classification (ICLR 2023 · similarity 0.88)
- Consistency Regularization for Variational Auto-Encoders (NeurIPS 2021 · similarity 0.88)
- A Contrastive Learning Approach for Training Variational Autoencoder Priors (NeurIPS 2021 · similarity 0.88)
- SeNM-VAE: Semi-Supervised Noise Modeling with Hierarchical Variational Autoencoder (CVPR 2024 · similarity 0.87)
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.