Research map / Probabilistic and scientific modeling / Variational and generative sampling
Decoupling Sparsity and Smoothness in the Dirichlet Variational Autoencoder Topic Model
Sophie Burkhardt, Stefan Kramer
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 |
|---|---|---|---|
| decouple sparsity and smoothness (DVAE Sparse vs DVAE) | DVAE · 20news (prior 0.02, 50 topics) | perplexity: 5009 -> 1066 | Improved |
| decouple sparsity and smoothness (DVAE Sparse vs DVAE) | DVAE · 20news (prior 0.02, 50 topics) | topic coherence: 0.324 -> 0.248 | Hurt |
| sample the KL-divergence during training instead of analytical (ablation) | DVAE · 20news (prior 0.02) | topic coherence: 0.33 -> 0.35 | Improved |
| sample the KL-divergence during training instead of analytical (ablation) | DVAE · 20news (prior 0.02) | final perplexity and KL | Improved |
| remove batch normalization (ablation) | Dirichlet VAEs · topic models | topic quality | Hurt |
| omit gradient correction term when shape augmentation is used (ablation) | DVAE · topic models | performance | No clear effect |
| RSVI Dirichlet reparameterization vs Laplace approximation (DVAE vs ProdLDA) | DVAE · 20news (prior 0.02, 50 topics) | topic coherence: 0.223 -> 0.324 | Improved |
| RSVI vs Weibull reparameterization | DVAE · 20news (prior 0.02, 50 topics) | topic coherence: 0.271 -> 0.324 | Improved |
| increase Dirichlet prior from 0.02 to 0.6 | DVAE · average coherence rank | average rank: 1.125 -> 1.750 | Hurt |
| Dirichlet prior (DVAE) vs Gaussian prior (NVDM) | topic model · 20news (prior 0.02, 50 topics) | topic coherence: 0.140 -> 0.324 | Improved |
| Dirichlet prior (DVAE) vs Gaussian prior (NVDM) | topic model · all datasets and topic counts (prior 0.02) | perplexity average rank: 7.125 vs 1.000 | Hurt |
Closest papers by meaning
- Autoencoding Variational Inference For Topic Models (ICLR 2017 · similarity 0.91)
- Sparse Autoencoders are Topic Models (ICML 2026 · similarity 0.89)
- Improving Variational Autoencoders with Density Gap-based Regularization (NeurIPS 2022 · similarity 0.88)
- An Instability in Variational Inference for Topic Models (ICML 2019 · similarity 0.88)
- Topic-VQ-VAE: Leveraging Latent Codebooks for Flexible Topic-Guided Document Generation (AAAI 2024 · similarity 0.86)
- Sparse Autoencoders, Again? (ICML 2025 · similarity 0.86)
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.
