Research map / Probabilistic and scientific modeling / Variational and generative sampling
Pólya Urn Latent Dirichlet Allocation: A Doubly Sparse Massively Parallel Sampler
Alexander Terenin, Måns Magnusson, Leif Jonsson, David Draper
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 |
|---|---|---|---|
| replace Dirichlet sampling of Phi with Poisson Polya urn | partially collapsed LDA · Enron/NYT/PubMed, K=100,1000 | runtime | Improved |
| replace Dirichlet sampling of Phi with Poisson Polya urn | partially collapsed LDA · Enron/NYT/PubMed, per iteration | log-posterior | No clear effect |
| replace Dirichlet with Poisson Polya urn at K=10 | partially collapsed LDA · Enron/NYT/PubMed K=10 | runtime and log-posterior | No clear effect |
| sample Phi with alias tables and Gaussian approximation | Phi sampling · Enron/NYT/PubMed | Phi sampling time: factor of four to eight | Improved |
| increase topics K from 100 to 1000 with Polya urn | z sampling · PubMed | z sampling time: by as much as 20% | Improved |
| use Polya Urn LDA instead of partially collapsed, sparse and light AD-LDA | parallel LDA samplers · NYT, K=100 and K=1000 | runtime | Improved |
| use Polya Urn LDA instead of partially collapsed LDA | partially collapsed LDA · NYT held-out | test log-likelihood and topic coherence | No clear effect |
| use Polya Urn LDA on single core vs sparse fully collapsed | fully collapsed sparse LDA · single core | runtime | Improved |
| increase vocabulary size / lower rare word threshold (ablation) | Polya Urn LDA vs partially collapsed · NYT | relative runtime | Improved |
| add more CPU cores | Polya Urn LDA · parallel scaling | runtime | Improved |
Closest papers by meaning
- A Reduction for Efficient LDA Topic Reconstruction (NeurIPS 2018 · similarity 0.89)
- Principled Selection of Hyperparameters in the Latent Dirichlet Allocation Model (JMLR 2018 · similarity 0.89)
- Attention-Based Sampler for Diffusion Language Models (NeurIPS 2026 · similarity 0.84)
- Discovering Sparsity Allocation for Layer-wise Pruning of Large Language Models (NeurIPS 2024 · similarity 0.84)
- Discrete Diffusion Models: Novel Analysis and New Sampler Guarantees (NeurIPS 2025 · similarity 0.84)
- How to Train Your Latent Diffusion Language Model Jointly With the Latent Space (NeurIPS 2026 · 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.
