3nd October, 5pm

Common Room

Presentations from current PhD students.

10th October, 5pm

Common Room

17th October, 5pm

Common Room

Presentations from new first year PhD students.

24th October, 5pm

Common Room

Francesca Basini

Title: Assessing competitive balance in the English Premier League for over forty seasons using a stochastic block model and some “behind the scenes”.

Abstract: In this talk, I am going to tell you how I ended up publishing a paper about football and why Tottenham Hotspur is my favourite team!
The main focus and motivation for this paper is to assess the issue of competitive balance in the context of the English First Division / Premier League over more than 40 seasons. We develop a Bayesian statistical network model based on an extension of the stochastic block model to measure the balance between teams in a league. We represent the outcome of all matches in a football season as a dense network with nodes identified by teams and categorical edges representing the outcome of each game as a win, draw, or a loss. We then infer the network block structure, that is the number of blocks in which the nodes are partitioned and their membership to a block, via efficient MCMC implementation of the allocation sampler.
Our analysis provides evidence suggesting a structural change around the early 2000s from a reasonably balanced league to a two-tier league.

31st October, 4:30pm

Common Room

Speaker: Alexander Kent

Title: A public introductory lecture on privacy.

Abstract: Given that the speaker for the first YRM introductory lecture is me, it should come as no surprise that the topic will be differential privacy. The main body of the talk will be an introduction and overview of common types of differential privacy, and we will explain some of the most fundamental building blocks of differentially private methods that are ubiquitous in its implementations, so that you too could develop private versions of your current research. We will then consider connections and applications of differential privacy in other fields, covering a range of active research topics in the department. Lastly, we will look at some real world implementations of differential privacy in the public sector and industry to both demonstrate the use of these tools, and to convince myself that my PhD has meaning.

7th November, 5pm

Common Room

Speaker: Kieran Drury

Title: Graphical Modelling of Complex Systems for Decision Support Systems with Limited Data

Abstract: This talk focuses on the application of graphical models such as Bayesian networks for the modelling of complex real-world systems for decision support purposes. In this way, we focus on the development of such graphical models in a causal setting such that we can intervene in the model in the same way as policies are designed to intervene in the corresponding real-world system. I will first give a general introduction to graphical models, before explaining how these models can be used to score each potential real-world policy to aid decision-making within a particular decision problem. Further, as data is often sufficiently sparse to prevent fully data-driven approaches to the construction of these models, we detail how expert judgements can be integrated into the model development process, and describe various methods for eliciting these judgements to counter various issues stemming from cognitive biases present even in the minds of world-leading domain experts. I will present two very different applications of this work, demonstrating two unique decision problems. The first is the application of this modelling to improving pollinator abundance in Australia, and the second is about choosing how and when police should intervene in suspected developing terrorism plots. Surely everyone is interested in either bees or terrorists, right?

14th November, 5pm

Common Room

Speaker: Alberto Bordino

Title: Tests of MCAR based on sample covariance matrices

Abstract: We study the problem of testing whether the missing values of a potentially high-dimensional dataset are Missing Completely at Random (MCAR). We relax the problem of testing MCAR to the problem of testing the compatibility of a sequence of covariance matrices, motivated by the fact that this procedure is feasible when the dimension grows with the sample size. Tests of compatibility can be used to test the feasibility of positive semi-definite matrix completion problems with noisy observations, and thus our results may be of independent interest. Our first contributions are to define a natural measure of the incompatibility of a sequence of correlation matrices, which can be characterised as the optimal value of a Semi-definite Programming (SDP) problem, and to establish a key duality result allowing its practical computation and interpretation. By studying the concentration properties of the natural plug-in estimator of this measure, we introduce novel hypothesis tests that we prove have power against all distributions with incompatible covariance matrices. The choice of critical values for our tests rely on a new concentration inequality for the Pearson sample correlation matrix, which may be of interest more widely. By considering key examples of missingness structures, we demonstrate that our procedures are minimax rate optimal in certain cases. We further validate our methodology with numerical simulations that provide evidence of validity and power, even when data are heavy tailed.

21st November, 5pm

Common Room

Speaker: David Huk

Title: From noise contrastive estimation to classifier-based copulas

Abstract: This talk will begin with an introduction to noise contrastive estimation and density ratio estimation by classification. We will cover the main ideas and methods available for these problems. We will then discuss copulas, starting from the definition and building intuition about their role for statistical modelling. Finally, I will present my recent work on classification-based copula density estimation which combines the aforementioned ideas into a new estimation procedure, yielding state-of-the-art performance among copula models.

28th November, 4:30pm

Common Room

Speaker: Andreea Popescu

Title: Epstein-Zin Stochastic Differential Utility

Abstract: Stochastic Differential Utility (SDU) is a non-linear generalisation of the classical time-additive expected utility framework. Intuitively, SDU allows for the separation between the risk aversion of an agent and their intertemporal substitutability, two characteristics that the in the classical von-Neuman Morgenstern expected utility setup are indistinguishable. Mathematically, the classical (concave, nondecreasing) utility function is generalised to an aggregator, and the stochastic differential utility process is defined in feedback form as the solution to some BSDE. The aim of this lecture is to introduce the Epstein-Zin SDU aggregator and to explore the classical investment-consumption problem in a continuous-time complete-market setting in the case when investors’ preferences are described by EZ SDU.

5th December, 5pm

Common Room

Small Talks