Research map / Probabilistic and scientific modeling / Variational and generative sampling
DVAE++: Discrete Variational Autoencoders with Overlapping Transformations
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 |
|---|---|---|---|
| Use overlapping transformations instead of Concrete | Nonlinear binary latent VAE · OMNIGLOT | log-likelihood: -116.69 -> -113.83 | Helped |
| Use overlapping transformations instead of Concrete | Nonlinear binary latent VAE · MNIST (static) | log-likelihood: -99.54 -> -99.16 | Helped |
| Use the marginal ELBO rather than the joint ELBO (ablation) | Overlapping-transformation VAE · Nonlinear OMNIGLOT | log-likelihood: -113.83 -> -113.76 | Helped |
| Use the marginal ELBO rather than the joint ELBO (ablation) | Overlapping-transformation VAE · Nonlinear MNIST (static) | log-likelihood: -95.04 -> -95.06 | Helped |
| Use an RBM global prior rather than a directed prior | VAE with overlapping transformations · MNIST (static), nonlinear, 2 latent layers | log-likelihood: -87.96 -> -84.27 | Helped |
| Replace normal local variables with Bernoulli variables (ablation) | DVAE++ · MNIST (static) | log-likelihood: -79.17 -> -79.72 | Hurt |
| Replace normal local variables with Bernoulli variables (ablation) | DVAE++ · MNIST (dynamic) | log-likelihood: -78.49 -> -79.55 | Hurt |
| Replace normal local variables with Bernoulli variables (ablation) | DVAE++ · OMNIGLOT | log-likelihood: -92.38 -> -93.95 | Hurt |
| Replace normal local variables with Bernoulli variables (ablation) | DVAE++ · Caltech-101 | log-likelihood: -81.88 -> -85.40 | Hurt |
| Replace normal local variables with Bernoulli variables (ablation) | DVAE++ · CIFAR10 | bits per dimension: 3.38 -> 3.59 | Hurt |
| Replace the RBM global prior with a Bernoulli prior (ablation) | DVAE++ with Bernoulli local variables · MNIST (static) | log-likelihood: -79.72 -> -79.90 | Hurt |
| Replace the RBM global prior with a Bernoulli prior (ablation) | DVAE++ with Bernoulli local variables · MNIST (dynamic) | log-likelihood: -79.55 -> -79.62 | Hurt |
Closest papers by meaning
- DVAE#: Discrete Variational Autoencoders with Relaxed Boltzmann Priors (NeurIPS 2018 · similarity 0.94)
- Discrete Variational Autoencoding via Policy Search (ICLR 2026 · similarity 0.91)
- Direct Optimization through $\arg \max$ for Discrete Variational Auto-Encoder (NeurIPS 2019 · similarity 0.89)
- InfoVAE: Balancing Learning and Inference in Variational Autoencoders (AAAI 2019 · similarity 0.89)
- Neural Discrete Representation Learning (NeurIPS 2017 · similarity 0.89)
- Discrete Variational Autoencoders (ICLR 2017 · similarity 0.89)
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.