Research map / Probabilistic and scientific modeling / Variational and generative sampling
Connections with Robust PCA and the Role of Emergent Sparsity in Variational Autoencoder Models
Bin Dai, Yu Wang, John Aston, Gang Hua, David Wipf
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 a stochastic encoder covariance (VAE instead of deterministic AE-l2) (ablation) | autoencoder on corrupted nonlinear manifold data · Synthetic manifold recovery with outliers | normalized MSE | Improved |
| use VAE instead of AE-l1 sparse latent penalty (ablation) | autoencoder on corrupted nonlinear manifold data · Synthetic manifold recovery with outliers | normalized MSE | Improved |
| use VAE instead of convex RPCA | manifold recovery with outliers · Synthetic manifold recovery with outliers | normalized MSE | Improved |
| add decoder hidden layers (Nd from 0 to 1) (ablation) | VAE with Ne=0 · Synthetic manifold, true latent dim 20 | nonzero columns in W1: 30.0 -> 21.1 | Improved |
| add decoder hidden layers (Nd from 0 to 3) (ablation) | VAE with Ne=3 · Synthetic manifold, true latent dim 20 | nonzero columns in W1: 30.0 -> 20.0 | Improved |
| add decoder hidden layers (Nd from 0 to 2) (ablation) | VAE with Ne=2 · Synthetic manifold, true latent dim 20 | nonzero columns in W1: 30.0 -> 20.0 | Improved |
| add encoder hidden layers (Ne from 0 to 3) at Nd=1 (ablation) | VAE · Synthetic manifold, true latent dim 20 | nonzero columns in W1: 21.1 -> 20.4 | Improved |
| use a VAE instead of an AE for latent dimension pruning on MNIST (ablation) | MNIST VAE with kappa above 15 · MNIST | nonzero columns in W1 | Improved |
| use tuned C2 = 10^3 for AE-l2 instead of default C2 = 1 (ablation) | AE-l2 · Synthetic manifold recovery with outliers | normalized MSE: C2 = 1 -> C2 = 10^3 | Improved |
| use 5 latent samples instead of 1 per input (ablation) | VAE · MNIST with outliers (appendix) | normalized MSE: tau = 1 -> tau = 5 | Improved |
| use VAE instead of convex RPCA | MNIST denoising · MNIST with outliers (appendix) | normalized MSE | Improved |
Closest papers by meaning
- Marginalization is not Marginal: No Bad VAE Local Minima when Learning Optimal Sparse Representations (ICML 2023 · similarity 0.92)
- Be a Goldfish: Forgetting Bad Conditioning in Sparse Linear Regression via Variational Autoencoders (ICML 2025 · similarity 0.92)
- From Variational to Deterministic Autoencoders (ICLR 2020 · similarity 0.92)
- Variational Autoencoders Pursue PCA Directions (by Accident) (CVPR 2019 · similarity 0.92)
- On the Value of Infinite Gradients in Variational Autoencoder Models (NeurIPS 2021 · similarity 0.91)
- Beyond Vanilla Variational Autoencoders: Detecting Posterior Collapse in Conditional and Hierarchical Variational Autoencoders (ICLR 2024 · similarity 0.90)
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.
