More Like this

1. Uncertainty Quantification for Bayesian Optimization

   Tuo, R.; Wang, W. (January 2022, Proceedings of Machine Learning Research)

   Bayesian optimization is a class of global optimization techniques. In Bayesian optimization, the underlying objective function is modeled as a realization of a Gaussian process. Although the Gaussian process assumption implies a random distribution of the Bayesian optimization outputs, quantification of this uncertainty is rarely studied in the literature. In this work, we propose a novel approach to assess the output uncertainty of Bayesian optimization algorithms, which proceeds by constructing confidence regions of the maximum point (or value) of the objective function. These regions can be computed efficiently, and their confidence levels are guaranteed by the uniform error bounds for sequential Gaussian process regression newly developed in the present work. Our theory provides a unified uncertainty quantification framework for all existing sequential sampling policies and stopping criteria. 

   more » « less
2. Distributionally Robust Optimization with Bias and Variance Reduction

   Mehta, Ronak; Roulet, Vincent; Pillutla, Krishna; Harchaoui, Zaid (May 2024, OpenReview)

   OpenReview (Ed.)

   We consider the distributionally robust optimization (DRO) problem with spectral risk-based uncertainty set and f-divergence penalty. This formulation includes common risk-sensitive learning objectives such as regularized condition value-at-risk (CVaR) and average top-k loss. We present Prospect, a stochastic gradient-based algorithm that only requires tuning a single learning rate hyperparameter, and prove that it enjoys linear convergence for smooth regularized losses. This contrasts with previous algorithms that either require tuning multiple hyperparameters or potentially fail to converge due to biased gradient estimates or inadequate regularization. Empirically, we show that Prospect can converge 2-3× faster than baselines such as stochastic gradient and stochastic saddle-point methods on distribution shift and fairness benchmarks spanning tabular, vision, and language domains. 

   more » « less
3. Hyperparameter Estimation in Bayesian MAP Estimation: Parameterizations and Consistency

   https://doi.org/10.5802/smai-jcm.62  

   Dunlop, Matthew M.; Helin, Tapio; Stuart, Andrew M. (April 2020, The SMAI journal of computational mathematics)

   null (Ed.)

   The Bayesian formulation of inverse problems is attractive for three primary reasons: it provides a clear modelling framework; it allows for principled learning of hyperparameters; and it can provide uncertainty quantification. The posterior distribution may in principle be sampled by means of MCMC or SMC methods, but for many problems it is computationally infeasible to do so. In this situation maximum a posteriori (MAP) estimators are often sought. Whilst these are relatively cheap to compute, and have an attractive variational formulation, a key drawback is their lack of invariance under change of parameterization; it is important to study MAP estimators, however, because they provide a link with classical optimization approaches to inverse problems and the Bayesian link may be used to improve upon classical optimization approaches. The lack of invariance of MAP estimators under change of parameterization is a particularly significant issue when hierarchical priors are employed to learn hyperparameters. In this paper we study the effect of the choice of parameterization on MAP estimators when a conditionally Gaussian hierarchical prior distribution is employed. Specifically we consider the centred parameterization, the natural parameterization in which the unknown state is solved for directly, and the noncentred parameterization, which works with a whitened Gaussian as the unknown state variable, and arises naturally when considering dimension-robust MCMC algorithms; MAP estimation is well-defined in the nonparametric setting only for the noncentred parameterization. However, we show that MAP estimates based on the noncentred parameterization are not consistent as estimators of hyperparameters; conversely, we show that limits of finite-dimensional centred MAP estimators are consistent as the dimension tends to infinity. We also consider empirical Bayesian hyperparameter estimation, show consistency of these estimates, and demonstrate that they are more robust with respect to noise than centred MAP estimates. An underpinning concept throughout is that hyperparameters may only be recovered up to measure equivalence, a well-known phenomenon in the context of the Ornstein–Uhlenbeck process. The applicability of the results is demonstrated concretely with the study of hierarchical Whittle–Matérn and ARD priors. 

   more » « less
4. Direct Synthesis of Iterative Algorithms With Bounds on Achievable Worst-Case Convergence Rate

   https://doi.org/10.23919/ACC45564.2020.9147401  

   Lessard, Laurent; Seiler, Peter (July 2020, American Control Conference)

   Iterative first-order methods such as gradient descent and its variants are widely used for solving optimization and machine learning problems. There has been recent interest in analytic or numerically efficient methods for computing worst-case performance bounds for such algorithms, for example over the class of strongly convex loss functions. A popular approach is to assume the algorithm has a fixed size (fixed dimension, or memory) and that its structure is parameterized by one or two hyperparameters, for example a learning rate and a momentum parameter. Then, a Lyapunov function is sought to certify robust stability and subsequent optimization can be performed to find optimal hyperparameter tunings. In the present work, we instead fix the constraints that characterize the loss function and apply techniques from robust control synthesis to directly search over algorithms. This approach yields stronger results than those previously available, since the bounds produced hold over algorithms with an arbitrary, but finite, amount of memory rather than just holding for algorithms with a prescribed structure. 

   more » « less
5. Four-dimensional mesospheric and lower thermospheric wind fields using Gaussian process regression on multistatic specular meteor radar observations

   https://doi.org/10.5194/amt-14-7199-2021  

   Volz, Ryan; Chau, Jorge L.; Erickson, Philip J.; Vierinen, Juha P.; Urco, J. Miguel; Clahsen, Matthias (January 2021, Atmospheric Measurement Techniques)

   Abstract. Mesoscale dynamics in the mesosphere and lower thermosphere (MLT) region have been difficult to study from either ground- or satellite-based observations. For understanding of atmospheric coupling processes, important spatial scales at these altitudes range between tens and hundreds of kilometers in the horizontal plane. To date, this scale size is challenging observationally, so structures are usually parameterized in global circulation models. The advent of multistatic specular meteor radar networks allows exploration of MLT mesoscale dynamics on these scales using an increased number of detections and a diversity of viewing angles inherent to multistatic networks. In this work, we introduce a four-dimensional wind field inversion method that makes use of Gaussian process regression (GPR), which is a nonparametric and Bayesian approach. The method takes measured projected wind velocities and prior distributions of the wind velocity as a function of space and time, specified by the user or estimated from the data, and produces posterior distributions for the wind velocity. Computation of the predictive posterior distribution is performed on sampled points of interest and is not necessarily regularly sampled. The main benefits of the GPR method include this non-gridded sampling, the built-in statistical uncertainty estimates, and the ability to horizontally resolve winds on relatively small scales. The performance of the GPR implementation has been evaluated on Monte Carlo simulations with known distributions using the same spatial and temporal sampling as 1 d of real meteor measurements. Based on the simulation results we find that the GPR implementation is robust, providing wind fields that are statistically unbiased with statistical variances that depend on the geometry and are proportional to the prior velocity variances. A conservative and fast approach can be straightforwardly implemented by employing overestimated prior variances and distances, while a more robust but computationally intensive approach can be implemented by employing training and fitting of model hyperparameters. The latter GPR approach has been applied to a 24 h dataset and shown to compare well to previously used homogeneous and gradient methods. Small-scale features have reasonably low statistical uncertainties, implying geophysical wind field horizontal structures as low as 20–50 km. We suggest that this GPR approach forms a suitable method for MLT regional and weather studies. 

   more » « less