Research map / Probabilistic and scientific modeling / Variational and generative sampling
Direct Optimization through $\arg \max$ for Discrete Variational Auto-Encoder
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 direct optimization instead of GSM | Discrete VAE · MNIST test | test loss: 167.88 -> 165.26 | Helped |
| Use direct optimization instead of the unbiased gradient method | Discrete VAE · Fashion-MNIST test | test loss: 200.88 -> 180.31 | Helped |
| Use direct optimization instead of GSM | Semi-supervised VAE · MNIST, 50-1200 labels | accuracy: 84.7% -> 92.6%; 92.7% -> 96.8% | Helped |
| Use direct optimization instead of GSM | Semi-supervised VAE · Fashion-MNIST, 50-1200 labels | accuracy: 61.2% -> 63.3%; 73.2% -> 73.7% | Helped |
| Add pairwise structured encoder interactions | Fashion-MNIST structured discrete VAE · Fashion-MNIST | test loss | Helped |
| Use a general pairwise encoder rather than a super-modular encoder | Fashion-MNIST structured discrete VAE · Fashion-MNIST | test loss | Helped |
| Add weak supervision using labeled examples | Semi-supervised VAE · MNIST and Fashion-MNIST | class accuracy | Helped |
| Use direct optimization instead of GSM | Discrete VAE · Fashion-MNIST test, k=10, 20, 30, 40, 50 | test loss: 238.37 -> 222.86; 211.87 -> 198.39; 197.01 -> 189.44; 195.22 -> 184.21; 191.00 -> 180.31 | Helped |
Closest papers by meaning
- Discrete Variational Autoencoding via Policy Search (ICLR 2026 · similarity 0.92)
- DVAE++: Discrete Variational Autoencoders with Overlapping Transformations (ICML 2018 · similarity 0.89)
- DVAE#: Discrete Variational Autoencoders with Relaxed Boltzmann Priors (NeurIPS 2018 · similarity 0.89)
- Variational Laplace Autoencoders (ICML 2019 · similarity 0.88)
- Be a Goldfish: Forgetting Bad Conditioning in Sparse Linear Regression via Variational Autoencoders (ICML 2025 · similarity 0.88)
- Beyond Vanilla Variational Autoencoders: Detecting Posterior Collapse in Conditional and Hierarchical Variational Autoencoders (ICLR 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.