arXiv subjects

Kellin Rumsey

Publications and source records attributed to Kellin Rumsey.

6 recordsLinked to original sources

arXiv2026

Ribbon: Scalable Approximation and Robust Uncertainty Quantification

Reliably quantifying predictive uncertainty is difficult for complex, high-dimensional, or misspecified models. Both fully Bayesian and bootstrap resampling methods provide principled uncertainty estimates but are often too expensive for modern machine-learning models because they require posterior sampling or repeated model refitting. We introduce Ribbon, a scalable approximation to Dirichlet-reweighted bootstrap uncertainty. Ribbon replaces repeated refitting with an influence-function linearization around a single fitted model, preserving the first-order data-reweighting structure of the Bayesian bootstrap while requiring only post-hoc linear algebra. Ribbon approximates the Bayesian-bootstrap or weighted-likelihood-bootstrap refitting target. With a general concentration parameter, Ribbon gives a calibrated Dirichlet-reweighting family whose uncertainty scale can be tuned on validation data. We show that Ribbon is asymptotically equivalent to a flat-prior Laplace approximation under correct likelihood specification and recovers the robust sandwich covariance under misspecification. Across synthetic regression, MNIST classification, and California Housing benchmarks, Ribbon provides competitive predictive performance and improved calibration in several settings while avoiding repeated model retraining.

stat.ML↗

arXiv2025

Fast Emulation, Modular Calibration, and Active Learning for Simulators with Functional Response

Scalable surrogate models enable efficient emulation of computer models (or simulators), particularly when dealing with large ensembles of runs. While Gaussian process (GP) models are commonly employed for emulation, they face limitations in scaling to large datasets. Furthermore, when dealing with dense functional output, such as spatial or time-series data, additional complexities arise, requiring careful handling to ensure fast emulation. This work presents a highly scalable emulator for functional data incorporating local Gaussian process regression. The emulator utilizes global GP lengthscale parameter estimates to scale the input space, leading to a substantial improvement in prediction speed. We demonstrate that our fast approximation-based emulator can serve as a viable alternative to a fully Bayesian approach for functional response, while drastically reducing computational costs. The proposed emulator is applied to quickly calibrate a multiphysics continuum hydrodynamics simulator with a large ensemble of 20000 runs. The methods presented are implemented in the R package FlaGP.

stat.ME↗

arXiv2025

A Partitioned Sparse Variational Gaussian Process for Fast, Distributed Spatial Modeling

The next generation of Department of Energy supercomputers will be capable of exascale computation. For these machines, far more computation will be possible than that which can be saved to disk. As a result, users will be unable to rely on post-hoc access to data for uncertainty quantification and other statistical analyses and there will be an urgent need for sophisticated machine learning algorithms which can be trained in situ. Algorithms deployed in this setting must be highly scalable, memory efficient and capable of handling data which is distributed across nodes as spatially contiguous partitions. One suitable approach involves fitting a sparse variational Gaussian process (SVGP) model independently and in parallel to each spatial partition. The resulting model is scalable, efficient and generally accurate, but produces the undesirable effect of constructing discontinuous response surfaces due to the disagreement between neighboring models at their shared boundary. In this paper, we extend this idea by allowing for a small amount of communication between neighboring spatial partitions which encourages better alignment of the local models, leading to smoother spatial predictions and a better fit in general. Due to our decentralized communication scheme, the proposed extension remains highly scalable and adds very little overhead in terms of computation (and none, in terms of memory). We demonstrate this Partitioned SVGP (PSVGP) approach for the Energy Exascale Earth System Model (E3SM) and compare the results to the independent SVGP case.

cs.LG↗

arXiv2024

Enhancing Approximate Modular Bayesian Inference by Emulating the Conditional Posterior

In modular Bayesian analyses, complex models are composed of distinct modules, each representing different aspects of the data or prior information. In this context, fully Bayesian approaches can sometimes lead to undesirable feedback between modules, compromising the integrity of the inference. This paper focuses on the "cut-distribution" which prevents unwanted influence between modules by "cutting" feedback. The multiple imputation (DS) algorithm is standard practice for approximating the cut-distribution, but it can be computationally intensive, especially when the number of imputations required is large. An enhanced method is proposed, the Emulating the Conditional Posterior (ECP) algorithm, which leverages emulation to increase the number of imputations. Through numerical experiment it is demonstrated that the ECP algorithm outperforms the traditional DS approach in terms of accuracy and computational efficiency, particularly when resources are constrained. It is also shown how the DS algorithm can be improved using ideas from design of experiments. This work also provides practical recommendations on algorithm choice based on the computational demands of sampling from the prior and cut-distributions.

stat.ME↗

arXiv2023

Generalized Bayesian MARS: Tools for Emulating Stochastic Computer Models

The multivariate adaptive regression spline (MARS) approach of Friedman (1991) and its Bayesian counterpart (Francom et al. 2018) are effective approaches for the emulation of computer models. The traditional assumption of Gaussian errors limits the usefulness of MARS, and many popular alternatives, when dealing with stochastic computer models. We propose a generalized Bayesian MARS (GBMARS) framework which admits the broad class of generalized hyperbolic distributions as the induced likelihood function. This allows us to develop tools for the emulation of stochastic simulators which are parsimonious, scalable, interpretable and require minimal tuning, while providing powerful predictive and uncertainty quantification capabilities. GBMARS is capable of robust regression with t distributions, quantile regression with asymmetric Laplace distributions and a general form of "Normal-Wald" regression in which the shape of the error distribution and the structure of the mean function are learned simultaneously. We demonstrate the effectiveness of GBMARS on various stochastic computer models and we show that it compares favorably to several popular alternatives.

stat.ME↗

arXiv2022

Bayesian Projection Pursuit Regression

In projection pursuit regression (PPR), an unknown response function is approximated by the sum of M "ridge functions," which are flexible functions of one-dimensional projections of a multivariate input space. Traditionally, optimization routines are used to estimate the projection directions and ridge functions via a sequential algorithm, and M is typically chosen via cross-validation. We introduce the first Bayesian version of PPR, which has the benefit of accurate uncertainty quantification. To learn the projection directions and ridge functions, we apply novel adaptations of methods used for the single ridge function case (M=1), called the Single Index Model, for which Bayesian implementations do exist; then use reversible jump MCMC to learn the number of ridge functions M. We evaluate the predictive ability of our model in 20 simulation scenarios and for 23 real datasets, in a bake-off against an array of state-of-the-art regression methods. Its effective performance indicates that Bayesian Projection Pursuit Regression is a valuable addition to the existing regression toolbox.

stat.ME↗