arXiv subjects

Simon Preston

Publications and source records attributed to Simon Preston.

8 recordsLinked to original sources

arXiv2025

Rolled Gaussian process models for curves on manifolds

Given a planar curve, imagine rolling a sphere along that curve without slipping or twisting, and by this means tracing out a curve on the sphere. It is well known that such a rolling operation induces a local isometry between the sphere and the plane so that the two curves uniquely determine each other, and moreover, the operation extends to a general class of manifolds in any dimension. We use rolling to construct an analogue of a Gaussian process on a manifold starting from a Euclidean Gaussian process with mean $m$ and covariance $K$, and refer to it as a rolled Gaussian process parameterized by $m$ and $K$. The resulting model is generative, and is amenable to statistical inference given data as curves on a manifold. We identify conditions on the manifold under which the rolling of $m$ equals the Fr\'echet mean of the rolled Gaussian process, propose computationally simple estimators of $m$ and $K$, and derive their rates of convergence. We illustrate with examples on the unit sphere, symmetric positive-definite matrices, and with a robotics application involving 3D orientations.

math.ST↗

arXiv2019

Invariance and identifiability issues for word embeddings

Word embeddings are commonly obtained as optimizers of a criterion function $f$ of a text corpus, but assessed on word-task performance using a different evaluation function $g$ of the test data. We contend that a possible source of disparity in performance on tasks is the incompatibility between classes of transformations that leave $f$ and $g$ invariant. In particular, word embeddings defined by $f$ are not unique; they are defined only up to a class of transformations to which $f$ is invariant, and this class is larger than the class to which $g$ is invariant. One implication of this is that the apparent superiority of one word embedding over another, as measured by word task performance, may largely be a consequence of the arbitrary elements selected from the respective solution sets. We provide a formal treatment of the above identifiability issue, present some numerical examples, and discuss possible resolutions.

stat.ML↗

arXiv2018

Classification and clustering for observations of event time data using non-homogeneous Poisson process models

Data of the form of event times arise in various applications. A simple model for such data is a non-homogeneous Poisson process (NHPP) which is specified by a rate function that depends on time. We consider the problem of having access to multiple independent observations of event time data, observed on a common interval, from which we wish to classify or cluster the observations according to their rate functions. Each rate function is unknown but assumed to belong to a finite number of rate functions each defining a distinct class. We model the rate functions using a spline basis expansion, the coefficients of which need to be estimated from data. The classification approach consists of using training data for which the class membership is known, to calculate maximum likelihood estimates of the coefficients for each group, then assigning test observations to a group by a maximum likelihood criterion. For clustering, by analogy to the Gaussian mixture model approach for Euclidean data, we consider mixtures of NHPP and use the expectation-maximisation algorithm to estimate the coefficients of the rate functions for the component models and group membership probabilities for each observation. The classification and clustering approaches perform well on both synthetic and real-world data sets. Code associated with this paper is available at https://github.com/duncan-barrack/NHPP .

cs.LG↗

arXiv2016

Nonparametric hypothesis testing for equality of means on the simplex

In the context of data that lie on the simplex, we investigate use of empirical and exponential empirical likelihood, and Hotelling and James statistics, to test the null hypothesis of equal population means based on two independent samples. We perform an extensive numerical study using data simulated from various distributions on the simplex. The results, taken together with practical considerations regarding implementation, support the use of bootstrap-calibrated James statistic.

stat.ME↗

arXiv2015

Bayesian model selection for the glacial-interglacial cycle

A prevailing viewpoint in palaeoclimate science is that a single palaeoclimate record contains insufficient information to discriminate between most competing explanatory models. Results we present here suggest the contrary. Using SMC^2 combined with novel Brownian bridge type proposals for the state trajectories, we show that even with relatively short time series it is possible to estimate Bayes factors to sufficient accuracy to be able to select between competing models. The results show that Monte Carlo methodology and computer power have now advanced to the point where a full Bayesian analysis for a wide class of conceptual climate models is now possible. The results also highlight a problem with estimating the chronology of the climate record prior to further statistical analysis, a practice which is common in palaeoclimate science. Using two datasets based on the same record but with different estimated chronologies results in conflicting conclusions about the importance of the orbital forcing on the glacial cycle, and about the internal dynamics generating the glacial cycle, even though the difference between the two estimated chronologies is consistent with dating uncertainty. This highlights a need for chronology estimation and other inferential questions to be addressed in a joint statistical procedure.

stat.AP↗

arXiv2015

AMP: a new time-frequency feature extraction method for intermittent time-series data

The characterisation of time-series data via their most salient features is extremely important in a range of machine learning task, not least of all with regards to classification and clustering. While there exist many feature extraction techniques suitable for non-intermittent time-series data, these approaches are not always appropriate for intermittent time-series data, where intermittency is characterized by constant values for large periods of time punctuated by sharp and transient increases or decreases in value. Motivated by this, we present aggregation, mode decomposition and projection (AMP) a feature extraction technique particularly suited to intermittent time-series data which contain time-frequency patterns. For our method all individual time-series within a set are combined to form a non-intermittent aggregate. This is decomposed into a set of components which represent the intrinsic time-frequency signals within the data set. Individual time-series can then be fit to these components to obtain a set of numerical features that represent their intrinsic time-frequency patterns. To demonstrate the effectiveness of AMP, we evaluate against the real word task of clustering intermittent time-series data. Using synthetically generated data we show that a clustering approach which uses the features derived from AMP significantly outperforms traditional clustering methods. Our technique is further exemplified on a real world data set where AMP can be used to discover groupings of individuals which correspond to real world sub-populations.

cs.LG↗

arXiv2015

Improved classification for compositional data using the $α$-transformation

In compositional data analysis an observation is a vector containing non-negative values, only the relative sizes of which are considered to be of interest. Without loss of generality, a compositional vector can be taken to be a vector of proportions that sum to one. Data of this type arise in many areas including geology, archaeology, biology, economics and political science. In this paper we investigate methods for classification of compositional data. Our approach centres on the idea of using the $α$-transformation to transform the data and then to classify the transformed data via regularised discriminant analysis and the k-nearest neighbours algorithm. Using the $α$-transformation generalises two rival approaches in compositional data analysis, one (when $α=1$) that treats the data as though they were Euclidean, ignoring the compositional constraint, and another (when $α=0$) that employs Aitchison's centred log-ratio transformation. A numerical study with several real datasets shows that whether using $α=1$ or $α=0$ gives better classification performance depends on the dataset, and moreover that using an intermediate value of $α$ can sometimes give better performance than using either 1 or 0.

stat.ME↗

arXiv2011

A data-based power transformation for compositional data

Compositional data analysis is carried out either by neglecting the compositional constraint and applying standard multivariate data analysis, or by transforming the data using the logs of the ratios of the components. In this work we examine a more general transformation which includes both approaches as special cases. It is a power transformation and involves a single parameter, α. The transformation has two equivalent versions. The first is the stay-in-the-simplex version, which is the power transformation as defined by Aitchison in 1986. The second version, which is a linear transformation of the power transformation, is a Box-Cox type transformation. We discuss a parametric way of estimating the value of α, which is maximization of its profile likelihood (assuming multivariate normality of the transformed data) and the equivalence between the two versions is exhibited. Other ways include maximization of the correct classification probability in discriminant analysis and maximization of the pseudo R-squared (as defined by Aitchison in 1986) in linear regression. We examine the relationship between the α-transformation, the raw data approach and the isometric log-ratio transformation. Furthermore, we also define a suitable family of metrics corresponding to the family of α-transformation and consider the corresponding family of Frechet means.

stat.ME↗