Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Online Bayesian Passive-Aggressive Learning
Tianlin Shi, Jun Zhu
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 |
|---|---|---|---|
| use online BayesPA mini-batch updates instead of batch MedLDA training | MedLDA / Gibbs MedLDA · 20NG | training time: more than an order of magnitude faster | Improved |
| use online BayesPA instead of batch multi-task MedLDA | MedLDAmt · Wiki 1.1M documents (multi-task) | training time at comparable F1: about two orders of magnitude faster | Improved |
| add per-mini-batch supervision (BayesPA) versus unsupervised online spLDA+SVM | online topic model · 20NG (K=80) | test error convergence | Improved |
| use online BayesPA instead of batch training | paMedLDA · 20NG | accuracy | No clear effect |
| use medium mini-batch size (64 or 256) rather than very small or large (ablation) | paMedLDA (K=40) · 20NG | convergence speed: |B| = 64 or 256 converge faster | Improved |
| increase optimization rounds I from 1 to 2 (ablation) | paMedLDAave and paMedHDPave · 20NG | accuracy: I = 1 suffers largely | Improved |
| increase optimization rounds I (ablation) | paMedLDAgibbs and paMedHDPgibbs · 20NG | accuracy | No clear effect |
| increase number of local samples J (ablation) | BayesPA topic models · 20NG | accuracy | No clear effect |
| increase local samples J from 1 to 5 (beta=0) (ablation) | paMedLDAgibbs · 20NG (K=40) | accuracy: 0.783 -> 0.808 | Improved |
| add burn-in steps with number of kept samples fixed (ablation) | paMedLDA · 20NG (K=40) | accuracy: (J, beta) = (3, 0), (5, 2), (9, 6) yield similar results | No clear effect |
| use averaging classifier instead of Gibbs classifier (ablation) | paMedLDA · 20NG | training speed | Improved |
| increase number of topics with one-pass training (ablation) | online Gibbs MedLDA · 20NG alt.atheism vs talk.religion.misc | accuracy | Hurt |
Closest papers by meaning
- Online Bayesian Moment Matching for Topic Modeling with Unknown Number of Topics (NeurIPS 2016 · similarity 0.85)
- BayesCNS: A Unified Bayesian Approach to Address Cold Start and Non-Stationarity in Search Systems at Scale (AAAI 2025 · similarity 0.84)
- Autoencoding Variational Inference For Topic Models (ICLR 2017 · similarity 0.84)
- Adaptive Elicitation of Latent Information Using Natural Language (ICML 2025 · similarity 0.83)
- Large Language Bayes (NeurIPS 2025 · similarity 0.82)
- Deep Bayesian Active Learning for Preference Modeling in Large Language Models (NeurIPS 2024 · similarity 0.82)
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.
