arXiv.itsnick.co

an independent field guide to 23 million papers from arXiv & OpenAlex — not affiliated with arXiv.org

Author

Luke Travis

4 papers · 2023–2024

2023: 32024: 1
2023peak 3 in 20232024

Frequent co-authors

Papers

showing 1–4 of 4 RSS

Semiparametric inference using fractional posteriors

Alice L'Huillier, Luke Travis, Ismaël Castillo +1 more

math.ST · · cited 2× · arXiv ↗ · more like this ·

We establish a general Bernstein--von Mises theorem for approximately linear semiparametric functionals of fractional posterior distributions based on nonparametric priors. This is illustrated in a number of nonparametric settings and for different classes of prior distributions, including Gaussian process priors. We show that fractional posterior credible sets can provide reliable semiparametric uncertainty quantification, but have inflated size. To remedy this, we further propose a \textit{shifted-and-rescaled} fractional posterior set that is an efficient confidence set having optimal size under regularity conditions. As part of our proofs, we also refine existing contraction rate results for fractional posteriors by sharpening the dependence of the rate on the fractional exponent.

A variational Bayes approach to inference for low-dimensional parameters in high-dimensional linear regression

Ismaël Castillo, Alice L'Huillier, Kolyan Ray +1 more

stat.ML · · cited 0× · arXiv ↗ · more like this ·

We propose a scalable variational Bayes method for statistical inference for a single or pre-specified low-dimensional subset of the coordinates of a high-dimensional parameter in sparse linear regression. Our approach relies on assigning a mean-field approximation to the nuisance coordinates and carefully modelling the conditional distribution of the target given the nuisance. This requires only a preprocessing step and preserves the computational advantages of mean-field variational Bayes, while ensuring accurate and reliable inference for the target parameter, including for uncertainty quantification. We investigate the numerical performance of our algorithm, showing that it performs competitively with existing methods. We further establish accompanying theoretical guarantees for estimation and uncertainty quantification in the form of a Bernstein--von Mises theorem for a possibly…

Pointwise uncertainty quantification for sparse variational Gaussian process regression with a Brownian motion prior

Luke Travis, Kolyan Ray

math.ST · · cited 0× · arXiv ↗ · more like this ·

We study pointwise estimation and uncertainty quantification for a sparse variational Gaussian process method with eigenvector inducing variables. For a rescaled Brownian motion prior, we derive theoretical guarantees and limitations for the frequentist size and coverage of pointwise credible sets. For sufficiently many inducing variables, we precisely characterize the asymptotic frequentist coverage, deducing when credible sets from this variational method are conservative and when overconfident/misleading. We numerically illustrate the applicability of our results and discuss connections with other common Gaussian process priors.

New Techniques and Methods for In-Situ Orbital Debris Detectors

Cornwell, Luke Travis

Artificial Intelligence · Kent Academic Repository (University of Kent) · · cited 0× · DOI ↗ · more like this ·

The rapidly increasing population of orbital debris in the near-Earth environment poses a significant hazard for operational spacecraft and future space missions. This has led to an increased need for in-situ detectors capable of observing, and distinguishing between, natural space dust and anthropogenic orbital debris, to both measure their flux and help quantify the threat that they pose. Accordingly, this thesis is concerned with the development of new techniques for in-situ orbital debris detection, with a focus on the development of acoustic thin film time of flight (TOF) detectors, specifically the Debris Resistive Acoustic Grid Orbital NASA-Navy Sensor (DRAGONS) developed by NASA, and my industrial partners AstroAcoustics. TOF detectors are valued for their impactor speed and direction measurement capabilities that allow the distinction between orbital debris and natural space…