# Tim Sullivan

Clear Search

### Computing with dense kernel matrices at near-linear cost in MMS

The paper “Compression, inversion, and approximate PCA of dense kernel matrices at near-linear computational complexity” by Florian Schäfer, Houman Owhadi, and myself has just appeared in print in Multiscale Modeling and Simulation. This paper shows how a surprisingly simple algorithm — the zero fill-in incomplete Cholesky factorisation — with respect to a cleverly-chosen sparsity pattern allows for near-linear complexity compression, inversion, and approximate PCA of square matrices of the form

$$\Theta = \begin{bmatrix} G(x_{1}, x_{1}) & \cdots & G(x_{1}, x_{N}) \\ \vdots & \ddots & \vdots \\ G(x_{N}, x_{1}) & \cdots & G(x_{N}, x_{N}) \end{bmatrix} \in \mathbb{R}^{N \times N} ,$$

where $$\{ x_{1}, \dots, x_{N} \} \subset \mathbb{R}^{d}$$ is a data set and $$G \colon \mathbb{R}^{d} \times \mathbb{R}^{d} \to \mathbb{R}$$ is a covariance kernel function. Such matrices play a key role in, for example, Gaussian process regression and RKHS-based machine learning techniques.

F. Schäfer, T. J. Sullivan, and H. Owhadi. “Compression, inversion, and approximate PCA of dense kernel matrices at near-linear computational complexity.” Multiscale Modeling & Simulation: A SIAM Interdisciplinary Journal 19(2):688–730, 2021. doi:10.1137/19M129526X

Abstract. Dense kernel matrices $$\Theta \in \mathbb{R}^{N \times N}$$ obtained from point evaluations of a covariance function $$G$$ at locations $$\{ x_{i} \}_{1 \leq i \leq N}$$ arise in statistics, machine learning, and numerical analysis. For covariance functions that are Green's functions of elliptic boundary value problems and homogeneously-distributed sampling points, we show how to identify a subset $$S \subset \{ 1 , \dots , N \}^2$$, with $$\# S = O ( N \log (N) \log^{d} ( N /\varepsilon ) )$$, such that the zero fill-in incomplete Cholesky factorisation of the sparse matrix $$\Theta_{ij} 1_{( i, j ) \in S}$$ is an $$\varepsilon$$-approximation of $$\Theta$$. This factorisation can provably be obtained in complexity $$O ( N \log( N ) \log^{d}( N /\varepsilon) )$$ in space and $$O ( N \log^{2}( N ) \log^{2d}( N /\varepsilon) )$$ in time; we further present numerical evidence that $$d$$ can be taken to be the intrinsic dimension of the data set rather than that of the ambient space. The algorithm only needs to know the spatial configuration of the $$x_{i}$$ and does not require an analytic representation of $$G$$. Furthermore, this factorization straightforwardly provides an approximate sparse PCA with optimal rate of convergence in the operator norm. Hence, by using only subsampling and the incomplete Cholesky factorization, we obtain, at nearly linear complexity, the compression, inversion and approximate PCA of a large class of covariance matrices. By inverting the order of the Cholesky factorization we also obtain a solver for elliptic PDE with complexity $$O ( N \log^{d}( N /\varepsilon) )$$ in space and $$O ( N \log^{2d}( N /\varepsilon) )$$ in time.

Published on Thursday 15 April 2021 at 12:00 UTC #publication #prob-num #schaefer #owhadi

### Convergence rates of Gaussian ODE filters in Statistics and Computing

The paper “Convergence rates of Gaussian ODE filters” by Hans Kersting, Philipp Hennig, and myself has just appeared in the journal Statistics and Computing. In this work, we examine the strong convergence rates of probabilistic solvers for ODEs of the form $$\dot{x}(t) = f(x(t))$$ that are based upon Gaussian filtering. In some sense, this work combines the numerical analysis perspective of Conrad et al. (2016) and Lie et al. (2019) with the filtering perspective on probabilistic numerical methods for ODEs of Schober et al. (2014).

H. Kersting, T. J. Sullivan, and P. Hennig. “Convergence rates of Gaussian ODE filters.” Statistics and Computing 30(6):1791–1816, 2020. doi:10.1007/s11222-020-09972-4

Abstract. A recently introduced class of probabilistic (uncertainty-aware) solvers for ordinary differential equations (ODEs) applies Gaussian (Kalman) filtering to initial value problems. These methods model the true solution $$x$$ and its first $$q$$ derivatives a priori as a Gauss–Markov process $$X$$, which is then iteratively conditioned on information about $$\dot{x}$$. This article establishes worst-case local convergence rates of order $$q + 1$$ for a wide range of versions of this Gaussian ODE filter, as well as global convergence rates of order $$q$$ in the case of $$q = 1$$ and an integrated Brownian motion prior, and analyses how inaccurate information on $$\dot{x}$$ coming from approximate evaluations of $$f$$ affects these rates. Moreover, we show that, in the globally convergent case, the posterior credible intervals are well calibrated in the sense that they globally contract at the same rate as the truncation error. We illustrate these theoretical results by numerical experiments which might indicate their generalizability to $$q \in \{ 2, 3 , \dots \}$$.

Published on Tuesday 15 September 2020 at 09:00 UTC #publication #stco #prob-num #kersting #hennig

### Adaptive reconstruction of monotone functions in special issue of Algorithms

The paper “Adaptive reconstruction of imperfectly-observed monotone functions, with applications to uncertainty quantification” by Luc Bonnet, Jean-Luc Akian, Éric Savin, and myself has just appeared in a special issue of the journal Algorithms devoted to Methods and Applications of Uncertainty Quantification in Engineering and Science. In this work, motivated by the computational needs of the optimal uncertainty quantification (OUQ) framework, we present and develop an algorithm for reconstructing a monotone function $$F$$ given the ability to interrogate $$F$$ pointwise but subject to partially controllable one-sided observational errors of the type that one would typically encounter if the observations would arise from a numerical optimisation routine.

L. Bonnet, J.-L. Akian, É. Savin, and T. J. Sullivan. “Adaptive reconstruction of imperfectly-observed monotone functions, with applications to uncertainty quantification.” Algorithms 13(8):196, 2020. doi:10.3390/a13080196

Abstract. Motivated by the desire to numerically calculate rigorous upper and lower bounds on deviation probabilities over large classes of probability distributions, we present an adaptive algorithm for the reconstruction of increasing real-valued functions. While this problem is similar to the classical statistical problem of isotonic regression, the optimisation setting alters several characteristics of the problem and opens natural algorithmic possibilities. We present our algorithm, establish sufficient conditions for convergence of the reconstruction to the ground truth, and apply the method to synthetic test cases and a real-world example of uncertainty quantification for aerodynamic design.

Published on Monday 17 August 2020 at 12:00 UTC #publication #algorithms #daad #ouq #isotonic #bonnet #akian #savin

### A rigorous theory of conditional mean embeddings in SIMODS

The article “A rigorous theory of conditional mean embeddings” by Ilja Klebanov, Ingmar Schuster, and myself has just appeared online in the SIAM Journal on Mathematics of Data Science. In this work we take a close mathematical look at the method of conditional mean embedding. In this approach to non-parametric inference, a random variable $$Y \sim \mathbb{P}_{Y}$$ in a set $$\mathcal{Y}$$ is represented by its kernel mean embedding, the reproducing kernel Hilbert space element

$$\displaystyle \mu_{Y} = \int_{\mathcal{Y}} \psi(y) \, \mathrm{d} \mathbb{P}_{Y} (y) \in \mathcal{G},$$

and conditioning with respect to an observation $$x$$ of a related random variable $$X \sim \mathbb{P}_{X}$$ in a set $$\mathcal{X}$$ with RKHS $$\mathcal{H}$$ is performed using the Woodbury formula

$$\displaystyle \mu_{Y|X = x} = \mu_Y + (C_{XX}^{\dagger} C_{XY})^\ast \, (\varphi(x) - \mu_X) .$$

Here $$\psi \colon \mathcal{Y} \to \mathcal{G}$$ and $$\varphi \colon \mathcal{X} \to \mathcal{H}$$ are the canonical feature maps and the $$C$$'s denote the appropriate centred (cross-)covariance operators of the embedded random variables $$\psi(Y)$$ in $$\mathcal{G}$$ and $$\varphi(X)$$ in $$\mathcal{H}$$.

Our article aims to provide rigorous mathematical foundations for this attractive but apparently naïve approach to conditional probability, and hence to Bayesian inference.

I. Klebanov, I. Schuster, and T. J. Sullivan. “A rigorous theory of conditional mean embeddings.” SIAM Journal on Mathematics of Data Science 2(3):583–606, 2020. doi:10.1137/19M1305069

Abstract. Conditional mean embeddings (CMEs) have proven themselves to be a powerful tool in many machine learning applications. They allow the efficient conditioning of probability distributions within the corresponding reproducing kernel Hilbert spaces by providing a linear-algebraic relation for the kernel mean embeddings of the respective joint and conditional probability distributions. Both centered and uncentered covariance operators have been used to define CMEs in the existing literature. In this paper, we develop a mathematically rigorous theory for both variants, discuss the merits and problems of each, and significantly weaken the conditions for applicability of CMEs. In the course of this, we demonstrate a beautiful connection to Gaussian conditioning in Hilbert spaces.

Published on Wednesday 15 July 2020 at 08:00 UTC #publication #simods #mathplus #tru2 #rkhs #mean-embedding #klebanov #schuster

### Optimality of probabilistic numerical methods

The paper “Optimality criteria for probabilistic numerical methods” by Chris Oates, Jon Cockayne, Dennis Prangle, Mark Girolami, and myself has just appeared in print:

C. J. Oates, J. Cockayne, D. Prangle, T. J. Sullivan, and M. Girolami. “Optimality criteria for probabilistic numerical methods” in Multivariate Algorithms and Information-Based Complexity, ed. F. J. Hickernell and P. Kritzer. Radon Series on Computational and Applied Mathematics 27:65–88, 2020. doi:10.1515/9783110635461-005

Abstract. It is well understood that Bayesian decision theory and average case analysis are essentially identical. However, if one is interested in performing uncertainty quantification for a numerical task, it can be argued that standard approaches from the decision-theoretic framework are neither appropriate nor sufficient. Instead, we consider a particular optimality criterion from Bayesian experimental design and study its implied optimal information in the numerical context. This information is demonstrated to differ, in general, from the information that would be used in an average-case-optimal numerical method. The explicit connection to Bayesian experimental design suggests several distinct regimes, in which optimal probabilistic numerical methods can be developed.

Published on Sunday 31 May 2020 at 08:00 UTC #publication #prob-num #oates #cockayne #prangle #girolami