Statistical Science


Statistical Science Seminars

A seminar series covering a broad range of applied and methodological topics in Statistical Science.

*** All talks will take place online until further notice ***

Usual time: Thursdays 14:00-15:00

Location: Zoom

Please email thomas dot bartlett dot 10 at ucl dot ac dot uk to join the mailing list, and receive the links to the talks.

Recent talks

Please subscribe to our Youtube channel, to view some recent talks from the series:


Upcoming talks

13 May 2021: Benjamin Eltzner (Universität Göttingen) - Testing for Uniqueness of Estimators

Uniqueness of the population descriptor is a standard assumption in asymptotic theory. However, so-called m-estimation problems, in which an estimator is determined by minimizing a cost function, often feature local minima of the sample cost function. These minima may stem from multiple global minima of the underlying population cost function. We present a hypothesis test to systematically determine for a given sample whether the underlying population cost function may have multiple global minima. We discuss three applications: 1) the mean on a non-euclidean data space, 2) non-linear regression and 3) Gaussian mixture clustering.

20 May 2021: Richard Samworth (Cambridge) - USP: an independence test that improves on Pearson's chi-squared 
and the G-test

We introduce the U-Statistic Permutation (USP) test of 
independence in the context of discrete data displayed in a contingency 
table. Either Pearson's chi-squared test of independence, or the 
Generalised Likelihood Ratio test (G-test), are typically used for this 
task, but we argue that these tests have serious deficiencies, both in 
terms of their inability to control the size of the test, and their power 
properties. By contrast, the USP test is guaranteed to control the size of 
the test at the nominal level for all sample sizes, has no issues with 
small (or zero) cell counts, and is able to detect distributions that 
violate independence in only a minimal way. The test statistic is derived 
from a U-statistic estimator of a natural population measure of 
dependence, and we prove that this is the unique minimum variance unbiased 
estimator of this population quantity.

In the last one-third of the talk, I will show how this is a special case 
of a much more general methodology and theory for independence testing.

27 May 2021: Matt Graham (UCL) - Lift and flow: manifold MCMC methods for efficient inference in stiff posteriors

A challenging regime for employing Markov chain Monte Carlo methods to perform Bayesian inference is when the observed data tightly constrains only some directions in the latent space. Such 'stiff' posterior distributions have varying scales across the latent space and can exhibit complex geometries which limit the performance of existing methods. In this talk I will present an approach for constructing efficient Markov kernels targeting such posteriors when the underlying generative model is differentiable. The posterior distribution is lifted on to a manifold embedded in a higher dimensional space, and a Hamiltonian flow on the manifold simulated to generate proposed moves. The lifted distribution remains diffuse in the presence of highly informative observations, allowing the flow to be simulated with large integrator steps and for chains to rapidly explore the lifted distribution. As we demonstrate empirically, this can lead to substantial improvements in sampling efficiency over competing approaches.

3 Jun 2021: Mariya Mamajiwala (UCL)


(Title and abstract TBC)

10 Jun 2021: Amanda Turner (Lancaster)


(Title and abstract TBC)

17 Jun 2021: Cédric Archambeau (Amazon)

(Title and abstract TBC)

24 Jun 2021: 16:00-17:00: Amy Willis (University of Washington)


(Title and abstract TBC)

1 July 2021: Takoua Jendoubi (UCL)


(Title and abstract TBC)

8 July 2021: Adrienne Leonard (Cirium / Pivigo)


(Title and abstract TBC)

15 July 2021: Magali Tournus (Penn State) - How to recover the fragmentation rate and kernel characterizing a size-structured population undergoing fragmentation?

We consider a suspension of particles that undergo fragmentation. We address the question of estimating the fragmentation parameters – i.e. the division rate B(x) and the fragmentation kernel k(y,x) – from measurements of the size particles distribution at various times. This is a natural question for any application where the sizes of the particles are measured experimentally whereas the fragmentation rates are unknown. The application that drives our work is the study of mechanical properties of amyloid fibrils that undergo fragmentation (are the mechanical properties related to toxicity?). In this talk, I will present the biological questions that motivate our work and the new experiments performed by Wei-Feng Xue team at Canterbury, then I will explain why the inverse problem is well posed under reasonable assumptions, and I will focus on how we can recover the fragmentation rate and kernel in practice.

Affiliated Seminars