Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
PASS-GLM: polynomial approximate sufficient statistics for scalable Bayesian GLM inference
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 |
|---|---|---|---|
| Use quadratic polynomial approximate sufficient statistics for logistic regression | PASS-LR2 · Criteo | negative test log-likelihood: 0.07 -> 0.045 | Helped |
| Use quadratic polynomial approximate sufficient statistics instead of SGD | logistic regression · Criteo | AUC | Hurt |
| Use quadratic polynomial approximate sufficient statistics instead of repeated-pass Laplace inference | logistic regression · ChemReact, Webspam, CovType, and CodRNA | runtime: an order of magnitude faster than SGD; 2--3 orders of magnitude faster than the Laplace approximation | Helped |
| Use quadratic polynomial approximate sufficient statistics | logistic regression · ChemReact, Webspam, and CovType | posterior mean estimates | Mixed |
| Use quadratic polynomial approximate sufficient statistics | logistic regression · ChemReact, Webspam, and CovType | posterior variance estimates | No clear effect |
| Use quadratic polynomial approximate sufficient statistics instead of MALA | logistic regression · ChemReact and Webspam | negative test log-likelihood: 10--100x longer for MALA to produce comparable results | No clear effect |
| Use quadratic polynomial approximate sufficient statistics instead of SGD | logistic regression · CovType | negative test log-likelihood: a tenth of the time | No clear effect |
| Use quadratic polynomial approximate sufficient statistics instead of MALA | logistic regression · CovType | negative test log-likelihood: 1000x longer for comparable results | No clear effect |
| Use quadratic polynomial approximate sufficient statistics | logistic regression · CodRNA | overall performance | Hurt |
| Stream inference over 40 million examples with PASS-LR2 instead of SGD | logistic regression · Criteo | runtime: about 10x slower | Hurt |
| Distribute PASS-LR2 computation across cores | PASS-LR2 · Criteo, 6M examples | speedup: K/2 | Helped |
Closest papers by meaning
- LR-GLM: High-Dimensional Bayesian Inference Using Low-Rank Data Approximations (ICML 2019 · similarity 0.93)
- Inflationary Flows: Calibrated Bayesian Inference with Diffusion-Based Models (NeurIPS 2024 · similarity 0.85)
- A Unified Confidence Sequence for Generalized Linear Models, with Applications to Bandits (NeurIPS 2024 · similarity 0.85)
- Turbocharging Gaussian Process Inference with Approximate Sketch-and-Project (NeurIPS 2025 · similarity 0.84)
- Finite-Sample Performance of Gradient Descent in Logistic Regression with Gaussian Design (NeurIPS 2026 · similarity 0.84)
- Least squares variational inference (NeurIPS 2025 · similarity 0.83)
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.