Research map / Probabilistic and scientific modeling / Variational and generative sampling
Autoencoding Variational Inference For Topic Models
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 LDA's word-level mixture with a product of experts | LDA, yielding ProdLDA · 20 Newsgroups, 50 topics | topic coherence: 0.11 -> 0.24 | Helped |
| Replace LDA's word-level mixture with a product of experts | LDA, yielding ProdLDA · 20 Newsgroups, 200 topics | topic coherence: 0.11 -> 0.19 | Helped |
| Replace LDA's word-level mixture with a product of experts | LDA, yielding ProdLDA · RCV1, 50 topics | topic coherence: 0.07 -> 0.14 | Helped |
| Replace LDA's word-level mixture with a product of experts | LDA, yielding ProdLDA · RCV1, 200 topics | topic coherence: 0.05 -> 0.12 | Helped |
| Use the inference network directly instead of optimizing on test documents | VAE-LDA · 20 Newsgroups test, 50 topics | perplexity: 1172 vs 1162 | No clear effect |
| Use the inference network directly instead of optimizing on test documents | VAE-LDA · 20 Newsgroups test, 200 topics | perplexity: 1168 vs 1151 | No clear effect |
| Use a Laplace approximation to the Dirichlet prior rather than a Gaussian prior (ablation) | ProdLDA with high learning rate · 20 Newsgroups, 50 topics | topic coherence: 0.08 -> 0.24 | Helped |
| Use high learning rate with batch normalization rather than low learning rate without batch normalization (ablation) | ProdLDA with Dirichlet prior · 20 Newsgroups, 50 topics | topic coherence: 0.016 -> 0.24 | Helped |
| Use high learning rate with batch normalization rather than low learning rate without batch normalization (ablation) | ProdLDA with Gaussian prior · 20 Newsgroups, 50 topics | topic coherence: 0.08 vs 0.08 | No clear effect |
| Use high-momentum, high-learning-rate training with batch normalization (ablation) | AVITM topic models · Topic modeling; numerical coherence reported for ProdLDA on 20 Newsgroups, 50 topics | topic coherence: 0.016 -> 0.24 | Helped |
Closest papers by meaning
- An Instability in Variational Inference for Topic Models (ICML 2019 · similarity 0.89)
- A Discrete Variational Recurrent Topic Model without the Reparametrization Trick (NeurIPS 2020 · similarity 0.88)
- WHAI: Weibull Hybrid Autoencoding Inference for Deep Topic Modeling (ICLR 2018 · similarity 0.88)
- Discovering Discrete Latent Topics with Neural Variational Inference (ICML 2017 · similarity 0.88)
- Latent Thought Models with Variational Bayes Inference-Time Computation (ICML 2025 · similarity 0.85)
- Topic-VQ-VAE: Leveraging Latent Codebooks for Flexible Topic-Guided Document Generation (AAAI 2024 · similarity 0.84)
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.