Several researchers have proposed minimisation of maximum mean discrepancy (MMD) as a method to quantise probability measures, i.e., to approximate a distribution by a representative point set. We consider sequential algorithms that greedily minimise MMD over a discrete candidate set. We propose a novel non-myopic algorithm and, in order to both improve statistical efficiency and reduce computational cost, we investigate a variant that applies this technique to a mini-batch of the candidate set at each iteration. When the candidate points are sampled from the target, the consistency of these new algorithms—and their mini-batch variants—is established. We demonstrate the algorithms on a range of important computational problems, including optimisation of nodes in Bayesian cubature and the thinning of Markov chain output.

The results of a series of theoretical studies are reported, examining the convergence rate for different approximate representations of α-stable distributions. Although they play a key role in modelling random processes with jumps and discontinuities, the use of α-stable distributions in inference often leads to analytically intractable problems. The LePage series, which is a probabilistic representation employed in this work, is used to transform an intractable, infinite-dimensional inference problem into a finite-dimensional (conditionally Gaussian) parametric problem. A major component of our approach is the approximation of the tail of this series by a Gaussian random variable. Standard statistical techniques, such as ExpectationMaximization (EM), Markov chain Monte Carlo, and Particle Filtering, can then be readily applied. In addition to the asymptotic normality of the tail of this series, we establish explicit, nonasymptotic bounds on the approximation error. Their proofs follow classical Fourier-analytic arguments, using Esséen's smoothing lemma. Specifically, we consider the distance between the distributions of: (i) the tail of the series and an appropriate Gaussian; (ii) the full series and the truncated series; and (iii) the full series and the truncated series with an added Gaussian term. In all three cases, sharp bounds are established, and the theoretical results are compared with the actual distances (computed numerically) in specific examples of symmetric αstable distributions. This analysis facilitates the selection of appropriate truncations in practice and offers theoretical guarantees for the accuracy of resulting estimates. One of the main conclusions obtained is that, for the purposes of inference, the use of a truncated series together with an approximately Gaussian error term has superior statistical properties and is likely a preferable choice in practice.

We report the results of several theoretical studies into the convergence rate for certain random series representations of α -stable random variables, which are motivated by and find application in modelling heavy-tailed noise in time series analysis, inference, and stochastic processes. The use of α -stable noise distributions generally leads to analytically intractable inference problems. The particular version of the Poisson series representation invoked here implies that the resulting distributions are “conditionally Gaussian,” for which inference is relatively straightforward, although an infinite series is still involved. Our approach is to approximate the residual (or “tail”) part of the series from some point, c > 0, say, to ∞, as a Gaussian random variable. Empirically, this approximation has been found to be very accurate for large c. We study the rate of convergence, as c → ∞, of this Gaussian approximation. This allows the selection of appropriate truncation parameters, so that a desired level of accuracy for the approximate model can be achieved. Explicit, nonasymptotic bounds are obtained for the Kolmogorov distance between the relevant distribution functions, through the application of probability-theoretic tools. The theoretical results obtained are found to be in very close agreement with numerical results obtained in earlier work.

In this paper we extend to the multidimensional case the modified Poisson series representation of linear stochastic processes driven by α-stable innovations. The latter has been recently introduced in the literature and it involves a Gaussian approximation of the residuals of the series, via the exact characterization of their moments. This allows for Bayesian techniques for parameter or state inference that would not be available otherwise, due to the lack of a closed-form likelihood function for the α-stable distribution. Simulation results are presented to validate the introduced extension and the quality of the approximation of the distribution. Finally, we show an example of generation from the process.