Aerosol-cloud interactions constitute the largest source of uncertainty in assessments of anthropogenic climate change. This uncertainty arises in part from the difficulty in measuring the vertical distributions of aerosols, and only sporadic vertically resolved observations are available. We often have to settle for less informative vertically aggregated proxies such as aerosol optical depth (AOD). In this work, we develop a framework for the vertical disaggregation of AOD into extinction profiles, that is, the measure of light extinction throughout an atmospheric column, using readily available vertically resolved meteorological predictors such as temperature, pressure, or relative humidity. Using Bayesian nonparametric modeling, we devise a simple Gaussian process prior over aerosol vertical profiles and update it with AOD observations to infer a distribution over vertical extinction profiles. To validate our approach, we use ECHAM-HAM aerosol-climate model data which offers self-consistent simulations of meteorological covariates, AOD, and extinction profiles. Our results show that, while very simple, our model is able to reconstruct realistic extinction profiles with well-calibrated uncertainty, outperforming by an order of magnitude the idealized baseline which is typically used in satellite AOD retrieval algorithms. In particular, the model demonstrates a faithful reconstruction of extinction patterns arising from aerosol water uptake in the boundary layer. Observations however suggest that other extinction patterns, due to aerosol mass concentration, particle size, and radiative properties, might be more challenging to capture and require additional vertically resolved predictors.

Simultaneous localization and mapping (SLAM) is the task of building a map representation of an unknown environment while at the same time using it for positioning. A probabilistic interpretation of the SLAM task allows for incorporating prior knowledge and for operation under uncertainty. Contrary to the common practice of computing point estimates of the system states, we capture the full posterior density through approximate Bayesian inference. This dynamic learning task falls under state estimation, where the state-of-the-art is in sequential Monte Carlo methods that tackle the forward filtering problem. In this paper, we introduce a framework for probabilistic SLAM using particle smoothing that does not only incorporate observed data in current state estimates, but it also backtracks the updated knowledge to correct for past drift and ambiguities in both the map and in the states. Our solution can efficiently handle both dense and sparse map representations by Rao-Blackwellization of conditionally linear and conditionally linearized models. We show through simulations and real-world experiments how the principles apply to radio (Bluetooth low-energy/Wi-Fi), magnetic field, and visual SLAM. The proposed solution is general, efficient, and works well under confounding noise.

Despite the growing availability of sensing and data in general, we remain unable to fully characterize many in-service engineering systems and structures from a purely data-driven approach. The vast data and resources available to capture human activity are unmatched in our engineered world, and, even in cases where data could be referred to as “big,” they will rarely hold information across operational windows or life spans. This paper pursues the combination of machine learning technology and physics-based reasoning to enhance our ability to make predictive models with limited data. By explicitly linking the physics-based view of stochastic processes with a data-based regression approach, a derivation path for a spectrum of possible Gaussian process models is introduced and used to highlight how and where different levels of expert knowledge of a system is likely best exploited. Each of the models highlighted in the spectrum have been explored in different ways across communities; novel examples in a structural assessment context here demonstrate how these approaches can significantly reduce reliance on expensive data collection. The increased interpretability of the models shown is another important consideration and benefit in this context.

We present a novel methodology for optimizing fiber optic network performance by determining the ideal values for attenuation, nonlinearity, and dispersion parameters in terms of achieved signal-to-noise ratio (SNR) gain from digital backpropagation (DBP). Our approach uses Gaussian process regression, a probabilistic machine learning technique, to create a computationally efficient model for mapping these parameters to the resulting SNR after applying DBP. We then use simplicial homology global optimization to find the parameter values that yield maximum SNR for the Gaussian process model within a set of a priori bounds. This approach optimizes the parameters in terms of the DBP gain at the receiver. We demonstrate the effectiveness of our method through simulation and experimental testing, achieving optimal estimates of the dispersion, nonlinearity, and attenuation parameters. Our approach also highlights the limitations of traditional one-at-a-time grid search methods and emphasizes the interpretability of the technique. This methodology has broad applications in engineering and can be used to optimize performance in various systems beyond optical networks.

Aeroengine performance is determined by temperature and pressure profiles along various axial stations within an engine. Given limited sensor measurements, we require a statistically principled approach for inferring these profiles. In this paper we detail a Bayesian methodology for interpolating the spatial temperature or pressure profile at axial stations within an aeroengine. The profile at any given axial station is represented as a spatial Gaussian random field on an annulus, with circumferential variations modelled using a Fourier basis and radial variations modelled with a squared exponential kernel. This Gaussian random field is extended to ingest data from multiple axial measurement planes, with the aim of transferring information across the planes. To facilitate this type of transfer learning, a novel planar covariance kernel is proposed. In the scenario where frequencies comprising the temperature field are unknown, we utilise a sparsity-promoting prior on the frequencies to encourage sparse representations. This easily extends to cases with multiple engine planes whilst accommodating frequency variations between the planes. The main quantity of interest, the spatial area average is readily obtained in closed form. We term this the Bayesian area average and demonstrate how this metric offers far more representative averages than a sector area average---a widely used area averaging approach. Furthermore, the Bayesian area average naturally decomposes the posterior uncertainty into terms characterising insufficient sampling and sensor measurement error respectively. This too provides a significant improvement over prior standard deviation based uncertainty breakdowns.

We apply deep kernel learning (DKL), which can be viewed as a combination of a Gaussian process (GP) and a deep neural network (DNN), to compression ignition engine emissions and compare its performance to a selection of other surrogate models on the same dataset. Surrogate models are a class of computationally cheaper alternatives to physics-based models. High-dimensional model representation (HDMR) is also briefly discussed and acts as a benchmark model for comparison. We apply the considered methods to a dataset, which was obtained from a compression ignition engine and includes as outputs soot and NOx emissions as functions of 14 engine operating condition variables. We combine a quasi-random global search with a conventional grid-optimization method in order to identify suitable values for several DKL hyperparameters, which include network architecture, kernel, and learning parameters. The performance of DKL, HDMR, plain GPs, and plain DNNs is compared in terms of the root mean squared error (RMSE) of the predictions as well as computational expense of training and evaluation. It is shown that DKL performs best in terms of RMSE in the predictions whilst maintaining the computational cost at a reasonable level, and DKL predictions are in good agreement with the experimental emissions data.

In Ludkovski, Risk, and Zail (2018), the email address for Jimmy Risk appeared incorrectly.

Jimmy Risk's email address should appear as [email protected].

The original article has been corrected to rectify this error.

We develop a Gaussian process (GP) framework for modeling mortality rates and mortality improvement factors. GP regression is a nonparametric, data-driven approach for determining the spatial dependence in mortality rates and jointly smoothing raw rates across dimensions, such as calendar year and age. The GP model quantifies uncertainty associated with smoothed historical experience and generates full stochastic trajectories for out-of-sample forecasts. Our framework is well suited for updating projections when newly available data arrives, and for dealing with “edge” issues where credibility is lower. We present a detailed analysis of GP model performance for US mortality experience based on the CDC (Center for Disease Control) datasets. We investigate the interaction between mean and residual modeling, Bayesian and non-Bayesian GP methodologies, accuracy of in-sample and out-of-sample forecasting, and stability of model parameters. We also document the general decline, along with strong age-dependency, in mortality improvement factors over the past few years, contrasting our findings with the Society of Actuaries (SOA) MP-2014 and -2015 models that do not fully reflect these recent trends.

Methods for efficient variance-based global sensitivity analysis of complex high-dimensional problems are presented and compared. Variance decomposition methods rank inputs according to Sobol indices that can be computationally expensive to evaluate. Main and interaction effect Sobol indices can be computed analytically in the Kennedy and O'Hagan framework with Gaussian processes. These methods use the high-dimensional model representation concept for variance decomposition that presents a unique model representation when inputs are uncorrelated. However, when the inputs are correlated, multiple model representations may be possible leading to ambiguous sensitivity ranking with Sobol indices. In this work, we present the effect of input correlation on sensitivity analysis and discuss the methods presented by Li and Rabitz in the context of Kennedy and O'Hagan's framework with Gaussian processes. Results are demonstrated on simulated and real problems for correlated and uncorrelated inputs and demonstrate the utility of variance decomposition methods for sensitivity analysis.

Radial Velocity Experiment (RAVE) observed ~500,000 southern sky stars between 2003 and 2013 in the infra-red calcium triplet (CaII) spectral region. In this study we extended the analysis of RAVE very metal-poor stars ([Fe/H] < −2) presented by Fulbright et al. (2010). We employed a novel method for identifying the metal-poor stars and developed a tool for modeling CaII lines where we also modeled the background noise to avoid systematical biases in the equivalent width (EW) measurements. Final metallicity values were derived with a flexible calibration approach using only 2MASS photometric data and EW measurements obtained from the RAVE spectra.

Fine regularity of stochastic processes is usually measured in a local way by localHölder exponents and in a global way by fractal dimensions. In the case of multiparameterGaussian random fields, Adler proved that these two concepts are connected under theassumption of increment stationarity property. The aim of this paper is to consider thecase of Gaussian fields without any stationarity condition. More precisely, we prove thatalmost surely the Hausdorff dimensions of the range and the graph in any ballB(t0,ρ)are bounded from above using the local Hölder exponent at t0. We definethe deterministic local sub-exponent of Gaussian processes, which allows to obtain analmost sure lower bound for these dimensions. Moreover, the Hausdorff dimensions of thesample path on an open interval are controlled almost surely by the minimum of the localexponents. Then, we apply these generic results to the cases of the set-indexed fractionalBrownian motion on RN, the multifractionalBrownian motion whose regularity function H is irregular and the generalized Weierstrassfunction, whose Hausdorff dimensions were unknown so far.

Consider a mean-reverting equation, generalized in the sense it is driven by a1-dimensional centeredGaussian process with Hölder continuous paths on [0,T] (T> 0). Taking thatequation in rough paths sense only gives local existence of the solution because thenon-explosion condition is not satisfied in general. Under natural assumptions, by usingspecific methods, we show the global existence and uniqueness of the solution, itsintegrability, the continuity and differentiability of the associated Itô map, and weprovide an Lp-convergingapproximation with a rate of convergence (p ≫ 1). The regularity of the Itô map ensures a largedeviation principle, and the existence of a density with respect to Lebesgue’s measure,for the solution of that generalized mean-reverting equation. Finally, we study ageneralized mean-reverting pharmacokinetic model.

In this paper, we describe techniques for evaluating the uncertainties associated with the assessment of form error, i.e., the departure from ideal geometry of a manufactured part, in coordinate metrology. The techniques take into account measurement uncertainty, sampling effects due to the fact that the form error is determined from a finite set of coordinate data points, and the spatial correlation of the form errors. The techniques are designed to be practical, without the need for complex computation.

Using integration by parts on Gaussian spacewe construct a Stein Unbiased Risk Estimator (SURE)for the drift of Gaussian processes, based on theirlocal and occupation times.By almost-sure minimization of the SURE risk ofshrinkage estimators we derive an estimation and de-noisingprocedure for an input signal perturbed by acontinuous-time Gaussian noise.

In this paper we deal with generalized fractional kinetic equations driven by a Gaussian noise, white in time and correlated in space, and where the diffusion operator is the composition of the Bessel and Riesz potentials for any fractional parameters. We give results on the existence and uniqueness of solutions by means of a weak formulation and study the Hölder continuity. Moreover, we prove the existence of a smooth density associated to the solution process and study the asymptotics of this density. Finally, when the diffusion coefficient is constant, we look for its Gaussian index.

We propose some construction of enhanced Gaussian processes usingKarhunen-Loeve expansion. We obtain a characterization and somecriterion of existence and uniqueness. Using rough-path theory, wederive some Wong-Zakai Theorem.

The center of the receptive field of some retinal ganglion cells exhibits an interesting fine structure: the relative amplitudes of responses to onset and responses to offset of a small spot of light varies systematically as the spot is positioned at various places within the center. Although this pattern may appear complicated, a simple model can account for it in detail. The model postulates that the ganglion cell receives input from separate ON and OFF processes within the center of its receptive field. These processes have the form of Gaussian functions and are laterally displaced from each other. These central ON and OFF input processes are not associated with the additional antagonistic surround of the receptive field.

The model is examined for various parameters of the input processes. The observed systematic variation in the ratio of offset to onset responses is predicted when the two processes are of nearly equal width (standard deviation of the Gaussians). Receptive fields made of more than two Gaussians produce various patterns, depending on the relative standard deviations of the Gaussians. Oblong fields, reminiscent of those found in visual cortex, may be generated from a relatively small number of circularly symmetric Gaussian processes.

Under regularity assumptions, we establish a sharp largedeviation principle for Hermitian quadratic forms ofstationary Gaussian processes. Our result is similar tothe well-known Bahadur-Rao theorem [2] on the samplemean. We also provide several examples of applicationsuch as the sharp large deviation properties ofthe Neyman-Pearson likelihood ratio test, of the sum of squares,of the Yule-Walkerestimator of the parameter of a stable autoregressive Gaussian process,and finally of the empirical spectral repartition function.

We generalise the work of Cramér and Leadbetter, Ylvisaker and Ito on the level crossings of a stationary Gaussian process to multivariate processes. Necessary and sufficient conditions for the existence of the expected number of crossings E(C) of the boundary of a region of ℝp by a stationary vector stochastic process are obtained, along with an explicit formula for E(C) in the Gaussian case. A rigorous proof of Belyaev's integral formula for the factorial moments of the number of exits from a region of ℝp is given for a class of processes which includes Gaussian processes having a finite second order spectral moment matrix. Applications to χ2 processes are briefly considered.

The ‘scanning process' Z(t), t ∈ ℝk of the title is a Gaussian random field obtained by associating with Z(t) the value of a set-indexed Brownian motion on the translate t + A0 of some ‘scanning set' A0. We study the basic properties of the random field Z relating, for example, its continuity and other sample path properties to the geometrical properties of A0. We ask if the set A0 determines the scanning process, and investigate when, and how, it is possible to recover the structure of A0 from realisations of the sample paths of the random field Z.