Abstract:SIAM/ASA Journal on Uncertainty Quantification, Volume 12, Issue 1, Page 69-100, March 2024. Variational inference is an approximation framework for Bayesian inference that seeks to improve quantified uncertainty in predictions by optimizing a simplified distribution over parameters to stand in for the full posterior. Capturing model variations that remain consistent with training data enables more robust predictions by reducing parameter sensitivity. This work introduces a fixed-point optimization for variational inference that is applicable when every feasible log density can be expressed as a linear combination of functions from a given basis. In such cases, the optimizer becomes a fixed-point of projective integral updates. When the basis spans univariate quadratics in each parameter, the feasible distributions are Gaussian mean-fields and the projective integral updates yield quasi-Newton variational Bayes (QNVB). Other bases and updates are also possible. Since these updates require high-dimensional integration, this work begins by proposing an efficient quasirandom sequence of quadratures for mean-field distributions. Each iterate of the sequence contains two evaluation points that combine to correctly integrate all univariate quadratic functions and, if the mean-field factors are symmetric, all univariate cubics. More importantly, averaging results over short subsequences achieves periodic exactness on a much larger space of multivariate polynomials of quadratic total degree. The corresponding variational updates require four loss evaluations with standard (not second-order) backpropagation to eliminate error terms from over half of all multivariate quadratic basis functions. This integration technique is motivated by first proposing stochastic blocked mean-field quadratures, which may be useful in other contexts. A PyTorch implementation of QNVB allows for better control over model uncertainty during training than competing methods. Experiments demonstrate superior generalizability for multiple learning problems and architectures.

Stochastic Collocation with Non-Gaussian Correlated Parameters via a New Quadrature Rule

Stochastic Collocation with Non-Gaussian Correlated Process Variations: Theory, Algorithms and Applications

Generation of Two Correlated Stationary Gaussian Processes

Adaptive Stochastic Collocation Method For Parameterized Statistical Timing Analysis With Quadratic Delay Model

A Simple Collocation-Type Approach to Numerical Stochastic Homogenization

Quadrangulation through morse-parameterization hybridization.

Efficient quadratures for high-dimensional Bayesian data assimilation

Accurately Solving the Poisson Equation by Combining Multiscale Radial Basis Functions and Gaussian Quadrature

Adaptive Stochastic Collocation Method (Ascm) For Parameterized Statistical Timing Analysis With Quadratic Delay Model

Uncertainty Quantification of RF Circuits Using Stochastic Collocation Techniques

Higher order QMC Galerkin discretization for parametric operator equations

Quasi-Monte Carlo Beyond Hardy-Krause

QMC integration for lognormal-parametric, elliptic PDEs: local supports and product weights

Gaussian quadrature rules for composite highly oscillatory integrals

A Stochastic Quasi-Newton Method in the Absence of Common Random Numbers

Smolyak Type Sparse Grid Collocation Method for Uncertainty Quantification of Nonlinear Stochastic Dynamic Equations

Conditional Bayesian Quadrature

High-order quasi-interpolation with generalized Gaussian kernels restricted over tori

Randomized matrix-free quadrature: unified and uniform bounds for stochastic Lanczos quadrature and the kernel polynomial method

Quadrature Methods for Highly Oscillatory Singular Integrals

Projective Integral Updates for High-Dimensional Variational Inference