Research map / Probabilistic and scientific modeling / Variational and generative sampling
Dispersed Exponential Family Mixture VAEs for Interpretable Text Generation
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 the dispersion term to the mixture VAE objective (ablation) | GM-VAE on PTB · PTB test | rPPL: 983.50 -> 257.68 | Helped |
| add the dispersion term to the mixture VAE objective (ablation) | GM-VAE on PTB · PTB test | BLEU: 2.34 -> 8.17 | Helped |
| add the dispersion term to the mixture VAE objective (ablation) | GM-VAE on DD · DD unsupervised | BLEU: 2.03 -> 7.63 | Helped |
| add the mutual information term (ablation) | DGM-VAE on PTB · PTB test | rPPL: 257.68 -> 247.37 | Helped |
| add the mutual information term (ablation) | DGM-VAE on PTB · PTB test | NLL: ≤104.26 -> ≤105.73 | Hurt |
| add the mutual information term (ablation) | DGM-VAE on DD · DD unsupervised | act homogeneity: 0.11 -> 0.23 | Helped |
| add the mutual information term (ablation) | DGM-VAE on DD · DD unsupervised | BLEU: 7.63 -> 7.39 | Hurt |
| add the mutual information term (ablation) | GM-VAE on PTB · PTB test | rPPL: 983.50 -> 287.07 | Helped |
| add the mutual information term (ablation) | GM-VAE on PTB · PTB test | BLEU: 2.34 -> 6.26 | Helped |
| replace independent discrete and continuous latents with a mixture prior linking the latent component to continuous z | DGM-VAE versus semi-VAE on PTB · PTB test | BLEU: 2.71 -> 8.17 | Helped |
| use a dispersion penalty rather than reweighting the whole KL term with β (ablation) | DGM-VAE versus β-GM-VAE on PTB · PTB test | BLEU: 8.17 versus 5.88; 5.92; 1.78 | Helped |
| use a dispersion penalty rather than aggressive posterior updates (ablation) | DGM-VAE versus GM-VAE + aggressive on PTB · PTB test | BLEU: 8.17 versus 1.47 | Helped |
Closest papers by meaning
- Toward Controlled Generation of Text (ICML 2017 · similarity 0.88)
- Improved Variational Autoencoders for Text Modeling using Dilated Convolutions (ICML 2017 · similarity 0.87)
- Improving Variational Autoencoders with Density Gap-based Regularization (NeurIPS 2022 · similarity 0.86)
- From Variational to Deterministic Autoencoders (ICLR 2020 · similarity 0.86)
- Topic-VQ-VAE: Leveraging Latent Codebooks for Flexible Topic-Guided Document Generation (AAAI 2024 · similarity 0.85)
- EQ-VAE: Equivariance Regularized Latent Space for Improved Generative Image Modeling (ICML 2025 · similarity 0.85)
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.