Research map / Probabilistic and scientific modeling / Variational and generative sampling
Stick-Breaking 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 |
|---|---|---|---|
| Replace Gaussian latent variables with Kumaraswamy stick-breaking latents | VAE · MNIST | -log p(x_i): 96.80 -> 98.01 | Hurt |
| Replace Gaussian latent variables with Kumaraswamy stick-breaking latents | VAE · MNIST+rot | -log p(x_i): 108.40 -> 112.33 | Hurt |
| Use Gauss-Logit rather than Kumaraswamy stick-breaking parametrization (ablation) | SB-VAE · MNIST and MNIST+rot | -log p(x_i): 98.01 -> 99.48; 112.33 -> 114.09 | Hurt |
| Use Gamma rather than Kumaraswamy stick-breaking parametrization (ablation) | SB-VAE · MNIST and MNIST+rot | -log p(x_i): 98.01 -> 100.74; 112.33 -> 113.22 | Hurt |
| Replace Gaussian latent variables with stick-breaking latent variables | Semi-supervised DGM · MNIST, 10% and 5% labels | test error (%): 3.95 -> 4.86; 4.74 -> 5.29 | Hurt |
| Replace Gaussian latent variables with stick-breaking latent variables | Semi-supervised DGM · MNIST, 1% labels | test error (%): 11.55 -> 7.34 | Helped |
| Replace Gaussian latent variables with stick-breaking latent variables | Semi-supervised DGM · MNIST+rot, 10%, 5%, and 1% labels | test error (%): 21.78 -> 11.78; 27.72 -> 14.27; 38.13 -> 27.67 | Helped |
| Replace Gaussian latent variables with stick-breaking latent variables | Semi-supervised DGM · SVHN, 10%, 5%, and 1% labels | test error (%): 36.08 -> 32.08; 48.75 -> 37.07; 69.58 -> 61.37 | Helped |
| Use an untruncated posterior, adding variational parameters online when more than 1% of the stick remains (ablation) | SB-VAE · Exploratory experiments | performance | No clear effect |
Closest papers by meaning
- Variational Autoencoders with Jointly Optimized Latent Dependency Structure (ICLR 2019 · similarity 0.90)
- Laplacian Autoencoders for Learning Stochastic Representations (NeurIPS 2022 · similarity 0.90)
- Shape your Space: A Gaussian Mixture Regularization Approach to Deterministic Autoencoders (NeurIPS 2021 · similarity 0.90)
- SQ-VAE: Variational Bayes on Discrete Representation with Self-annealed Stochastic Quantization (ICML 2022 · similarity 0.89)
- Variational Learning of Fractional Posteriors (ICML 2025 · similarity 0.89)
- Beyond Vanilla Variational Autoencoders: Detecting Posterior Collapse in Conditional and Hierarchical Variational Autoencoders (ICLR 2024 · similarity 0.88)
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.