Research map / Probabilistic and scientific modeling / Variational and generative sampling
Greedy inference with structure-exploiting lazy maps
Michael Brennan, Daniele Bigoni, Olivier Zahm, Alessio Spantini, Youssef Marzouk
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 a computed-basis lazy IAF instead of a baseline IAF | Parkinson's logistic regression with 20 observations · Parkinson's logistic regression (20 observations) | variance diagnostic: 23.8 (2.59) -> 7.68 (2.07) | Improved |
| Restrict the computed-basis lazy IAF to rank 20 rather than rank 500 | Parkinson's logistic regression with 20 observations · Parkinson's logistic regression (20 observations) | variance diagnostic: 7.68 (2.07) -> 1.58 (0.433) | Improved |
| Greedily compose 3 rank-200 lazy layers rather than use a baseline IAF | Parkinson's logistic regression with 605 observations · Parkinson's logistic regression (605 observations) | trace diagnostic Hᴮ/2: 980 (84.3) -> 526 (30.6) | Improved |
| Use a computed-basis U-IAF rather than a baseline IAF | Parkinson's logistic regression with 605 observations · Parkinson's logistic regression (605 observations) | Δ ELBO: -- -> 27.6 (0.529) | Improved |
| Greedily compose 3 lazy affine layers rather than use a baseline affine map | Yacht hydrodynamics Bayesian neural network · Yacht hydrodynamics Bayesian neural network | variance diagnostic: 1.6e4 (5.8e4) -> 97.5 (6.47) | Improved |
| Use the learned U-IAF transport to precondition HMC instead of sampling the target directly | Parkinson's logistic regression with 605 observations · Parkinson's logistic regression (605 observations) | average component-wise ESS: 0.065% -> 0.87% | Improved |
| Use the learned G3-Affine transport to precondition HMC instead of sampling the target directly | Yacht hydrodynamics Bayesian neural network · Yacht hydrodynamics Bayesian neural network | average component-wise ESS: 0.06% -> 0.56% | Improved |
| Use the greedy lazy transport to precondition pCN instead of applying pCN directly | High-dimensional elliptic PDE inverse problem · Elliptic PDE inverse problem (2601 dimensions) | acceptance rate: under 0.4% -> 28.2% | Improved |
Closest papers by meaning
- Inference via Low-Dimensional Couplings (JMLR 2018 · similarity 0.89)
- LazyDINO: Fast, Scalable, and Efficiently Amortized Bayesian Inversion via Structure-Exploiting and Surrogate-Driven Measure Transport (JMLR 2026 · similarity 0.88)
- Model-Informed Flows for Bayesian Inference (NeurIPS 2025 · similarity 0.85)
- Provable Posterior Sampling with Denoising Oracles via Tilted Transport (NeurIPS 2024 · similarity 0.85)
- Inflationary Flows: Calibrated Bayesian Inference with Diffusion-Based Models (NeurIPS 2024 · similarity 0.84)
- Generalized Variational Inference via Optimal Transport (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.
