Back to top

Previous Graduate Lectures

21 Sep 20

Previous Graduate Lectures:

TRINITY TERM 2020

Speaker:   Dominic Yeo, Department of Statistics

Title:          An Introduction to Random Graphs

Third Year Lightning Talks
Speaker:   Anja Sturm, University of Göttingen

Title:           Interacting particle systems:From local stochastic interactions to global phenomena

Abstract:  Stochastic interacting particle systems describe the time evolution of particle systems that are distributed on a discrete space (such as the integer lattice). The random changes to these systems are local, and one of the main points  of interest is to understand the global and long-time behavior that arises from a particular set of local interaction rules.

In this talk, we will introduce some classical interacting particle systems including the contact and the voter model. We will relate them to non-spatial counterparts, in particular Galton-Watson branching processes and Cannings models, and discuss properties such as long-time survival and the existence of (nontrivial) invariant laws.

We will then consider these properties for a particular one dimensional model called the cooperative branching coalescent. Here, particles perform independent random walks, only pairs of particles occupying neighboring sites can produce new particles (cooperative branching) and particles that land on an occupied site merge with the particle present on that site (coalescence).

We show that this system undergoes a phase transition as the branching rate is increased regarding its survival and upper invariant laws. These results were obtained in joint work with  Jan Swart (UTIA Prag).

Speaker:   Graham Lee, Research Software Engineer, Oxford RSE Group

Title:          Developing better code with automated testing

Abstract:  If we want reliable, reproducible simulations and data analysis software, we need to know that we have implemented our code correctly. Further, we need to be confident that changes we make to the code do not introduce unintended flaws. Automated testing is a technique widely used in industry to capture information about the expected behaviour of software and ensure that the system retains that behaviour through its evolution. In this talk, Graham explores the application of the technique to scientific software.

Distinguished Speaker Seminar

Speaker:    Nick Jewell, University of California, Berkeley School of Public Health

Title:           Cluster-Randomised Test Negative Designs: Inference and Application to Vector Trials to Eliminate Dengue

Abstract:   The successful introduction of the intracellular bacterium Wolbachia into Aedes aegypti mosquitoes enables a practical approach for dengue prevention through release of Wolbachia-infected mosquitoes. Wolbachia reduces dengue virus replication in the mosquito and, once established in the mosquito population, it is possible that this will provide a long-term and sustainable approach to reducing or eliminating dengue transmission. A critical next step is to assess the efficacy of Wolbachia deployments in reducing dengue virus transmission in the field.  I will  describe and discuss the statistical design of a large-scale cluster randomised test-negative parallel arm study to measure the efficacy of such interventions. Comparison of permutation inferential approaches to model based methods will be described. Extensions to allow for individual covariates, and alternate designs such as the stepped wedge approach, will also be briefly introduced. There are also interesting questions regarding interrupted time-series methods associated with analysing pilot site data.

Speaker:    Cora Mezger, Department of Statistics

Title:           Opportunities for using machine learning in official statistics

First Year Poster Session
Collaborative Careers Event with Medical Sciences

HILARY TERM 2020

Corcoran Memorial Prize Award and Lecture

Speaker: Professor Frank den Hollander, Leiden University

Title:        Synchronisation with noise

Followed by a short talk from the Corcoran Prize Winner, Dr Sarah Penington

Speaker: David Kell, CTO, Gyana Ltd

Title:       Making the Transition from Academia to Startup

Machine Learning Talk:

Speaker:  Nicolas Hees and Leonard Hasenclever

Title:        Deep reinforcement learning for control – algorithms and architectures

Speaker: Jotun Hein, Department of Statistics

Title:       Statistical Biological Sequence Analysis

Machine Learning Talk:

Speaker:   Arthur Gretton

Title:         Some results and conjectures for GAN critic design

Machine Learning Talk:

Speaker:   Razvan Pascanu

Title:         Optimization, learning and data-efficiency

Abstract:  Throughout this lecture I would like to ask the question of how we can improve training efficiency for neural networks. This question had been asked many times, from many angles. The angle we will adopt here is to look at the specific role that the learning algorithm or optimizer plays in determining training efficiency. The lecture will start with a brief overview of learning techniques, going from stochastic gradient descent to second order methods and natural gradient. Each method will be introduced and discussed, highlighting both its strengths and weaknesses. We will follow, afterwards, with a discussion on a more recent paradigm for efficient learning, namely meta-learning. The topic has seen an extensive growth in the last few years and we will not extensively cover all different forms of meta-learning, but rather focus on those that explicitly learn or parameterize an optimizer. More specifically, we will present gradient based methods such as meta-gradients and meta-curvature, Warp Gradient Descent, and compare them with other families of meta-learning algorithms. For simplicity, a large part of the discussion will be carried out in the context of supervised learning. However we will also discuss the RL setting, particularly for meta-learning. We will describe the unique challenges imposed by RL and how these could affect the presented algorithms. Finally we will draw some conclusions regarding the role of the optimizer in improving efficiency of learning and discuss potential directions for moving forward.

Machine Learning Talk:

Speaker: Silvia Chiappa

 

 

MICHAELMAS TERM 2019

Active and Passive Presentations: how to present your work, and how to get the most out of seminars, lectures and poster sessions. Charlotte Deane and Gesine Reinert, Department of Statistics
How to learn human values and solve all ethical problems – adequately  Stuart Armstrong, Future of Humanity Institute, University of Oxford

Bio: Stuart’s research at the Future of Humanity Institute centers on the safety and possibilities of Artificial Intelligence (AI), how to define the potential goals of AI and map humanity’s partially defined values into it, and the long term potential for intelligent life across the reachable universe. He has been working with people at FHI and other organizations, such as DeepMind, to formalize AI desiderata in general models so that AI designers can include these safety methods in their designs. His collaboration with DeepMind on “Interruptibility” has been mentioned in over 100 media articles.

How to have a successful internship during your DPhil Clare West, Susan Leung and Lyuba Bozhilova

Bio: Clare West, Susan Leung, and Lyuba Bozhilova are fourth year SABS CDT students and members of the Oxford Protein Informatics Group, here in the Department of Statistics, who have all had successful internships.

Clare worked for three months in Westminster at the Parliamentary Office of Science and Technology (POST), see https://www.blopig.com/blog/2019/03/not-proteins-in-parliament/.

Susan spent three months working on a Google Summer of Code project with Dr Greg Landrum, lead developer of the widely-used open source cheminformatics toolkit, RDKit; Susan worked on software development in C++, and MolVS, which provides very useful functionality for molecular validation and standardization; see https://summerofcode.withgoogle.com/archive/2018/projects/4878514637504512/.

Lyuba spent two and a half months working for BenevolentAI, a biotech company based in London; see https://benevolent.ai/. She split her time between data engineering and precision medicine research.

A few examples of how statistics is used when phasing and imputing with sequencing reads in genetics Robbie Davies, Department of Statistics

Bio: Robbie is a new Associate Professor in the Department of Statistics, having started in July 2019. Robbie is a graduate of the GMS program, completing his degree in 2015. His research is in statistical genetics, and often, but not always, related to humans and with a medical focus or purpose.

Student Entrepreneurs Cath Spence, Principal Licensing & Ventures Manager – Incubator Lead

Short description:  Have you ever wondered if you have what it takes? If you have the spark?  If you would like to explore what it takes to be an entrepreneur, then this your chance.

Following a hugely successful pilot programme in summer 2019, we are planning StEP ‘Ignite’. It’s a 4 week programme running over the holiday periods, 06-17 January 2020 and 16-27 March 2020.  StEP ‘Ignite’ has been developed by Oxford University Innovation (OUI) and supported by Oxford Sciences Innovation (OSI), and The Oxford Foundry.

OUI will make University IP available to Oxford University student groups for the purpose of putting together investable business cases. The groups will receive: a stipend (£1,500 for applicants who can commit to the full-time programme), a place to work, mentoring, free access to an intensive training programme and unlimited coffee! At the end all groups will have the chance to pitch for £25,000 and the chance to work with OSI on putting a more substantial first round investment together to start their new spinout.

Hidden relatedness, natural selection and disease heritability in the human genome Pier Palamara, Department of Statistics

Bio: Pier is an associate professor at the Department of Statistics in the University of Oxford. He is broadly interested in developing new computational methods to solve problems in population and medical genetics. Before coming to Oxford he spent a few years at the Harvard Chan School of Public Health and at the Broad Institute of MIT and Harvard. He received my PhD in computer science from Columbia University. His early research and training were in artificial intelligence and cognitive robotics.

HILARY TERM 2019

Speaker:        Kevin Sharp, Department of Statistics

Title:               Multiple phenotype models and the genetic basis of brain Structure and function

Speaker:        Abdel Turkmani, Util

Title:  Util – using machine learning to help investors make a positive impact on people, planet and pocket.

Speaker:  Christina Goldschmidt, Department of Statistics

Title: Exchangeability

Speaker:       Verena Heise, Nuffield Department of Population Health

Title:   Robust Research – Why do we care?

Speaker:  Reuben Binns, Department of Computer Science

Title:  Fair machine learning: how can data-driven decision-making  perpetuate bias, and what should be done about it? 

2nd year poster session

Week 8 Thursday 7th March:  Blenheim Chalcot

Title:  Industry Talk

HILARY TERM 2018

Speaker:  Jessie Wu, Department of Statistics, University of Oxford

Title:          Dirichlet process models and their applications

Abstract:   This is an introductory lecture to Dirichlet Processes (DP). The beginning of the lecture will introduce the definition and the various representations of DP to demonstrate how the process can be constructed.  Popular variations of DP will also be presented along with the applications of these models, such as clustering and nonparametric regression.

Speaker:  Ben  Bloem-Reddy, Department of Statistics, University of Oxford

Title:         An introduction to exchangeable random partitions and random discrete probability measures

Abstract: Random partition processes underly many constructions for Bayesian models of clustered data; the Chinese Restaurant Process (CRP) is the prototypical example. They are also perhaps the simplest type of combinatorial stochastic process, and are a starting point for studying processes with more complex structure. In the first part of the lecture, I will give an introduction to exchangeable random partitions and their connection to random discrete probability measures via Kingman’s paintbox, with special focus on the CRP. In the second part, I will focus on the properties of the Dirichlet Process (DP), and how different constructions of the DP lead to various classes of random probability measures that appear in the probability and Bayesian nonparametrics literature. Some of these properties will be used in the subsequent week’s grad lecture by Jessie Wu on applications of and inference with models that use the DP as a building block.

Speaker:   Marco Scutari, Department of Statistics, University of Oxford

Title:           What is new in R?

Abstract:  The capabilities of R have evolved considerably since its first public releases in the 1990s, with each major version bringing in new functionality. In this lecture I will discuss two areas which have seen major evolutions in R 3.x.y: the inclusion in the core distribution of the “parallel” package , which provides facilities for parallel computing; and the release of the Hadleyverse packages, which streamline common data cleaning tasks.

Speaker:  Jotun Hein, Department of Statistics, University of Oxford

Title:        Statistical Alignment with Long Insertion-Deletions for 2 and More Sequences

Speaker:  Mareli Grady, Department of Statistics

Title:         Hands-On Statistics: Getting started in Outreach and Public Engagement

Abstract:  The benefits of engaging in outreach and public engagement are numerous and the impacts important. In this talk we will explore the opportunities available in the Department and at Oxford in general, show some examples of outreach and public engagement activities currently in use and give you the opportunity to design and discuss some ideas for yourself. Put your creative hat on and join in!

MICHAELMAS TERM 2017:

Speaker:  Jen Rogers, Department of Statistics, University of Oxford

Abstract:  Jen took on the role of Director of Statistical Consultancy Services within the Department in July last year. In this talk she will be presenting her experiences of the job, talking about what is like to work with industry on a consultancy basis and professional aspects associated with the role. She will go through case studies of work that she has carried out, introducing the kinds of statistical problems that she typically encounters. Finally, she will also be outlining ways in which you may be able to get involved with consultancy activities within the Department.

Speaker: Dr Ricardo Silva, Lecturer in Statistics, UCL

Abstract: In this exposition, we will discuss the common tools used in the machine learning community to describe causal assumptions and how this leads to particular ways of thinking concerning the estimation of causal effects. We will focus on two main case studies: how to combine data from observational and experimental studies; and how to criticise ways of adjusting for confounding in observational studies, given that background knowledge my be imperfect and hide default assumptions with unintended consequences. 

Date:        Friday 27th October, 12.00 noon

Abstract:  XTX Markets trades an average of $80B daily in thousands of financial instruments  on an electronic basis with  little human interaction. This adds liquidity to the global financial markets and is made possible by taking a highly systematic, purely data-driven, approach to trading.  In this talk, we will describe how trading can be automated, the sort of problems that it poses, and how a small team of mathematicians, statisticians and computer scientists tackles them.

Graduate Talk  by XTX Markets followed by pizza lunch.

Speaker: Geoff Nicholls, Department of Statistics, University of Oxford

Title:       The Savage Axioms for Dummies

Abstract: In Bayesian inference the subjective expected utility is an object of foundational importance. The decision maker’s choices maximise this utility, and so it decides some elements of our statistical methodology.
For the SEU to exist we need a prior and a utility representing the subjective beliefs of the analyst to exist.

The Savage Axioms are a set of axioms which impose a certain “consistency” on our beliefs and values, and are sufficient for a representative prior and utility function to exist. I will present the axioms in the form given by Maurice de Groot,and look at two famous paradoxes which show, or appear to show, that some very natural human beliefs and values are in conflict with these axioms.I will mention some approaches to resolving of these paradoxes and conclude with a brief evaluation of the axioms.

Speakers: Professor Charlotte Deane and Professor Gesine Reinert, Department of Statistics, University of Oxford

Title:         Active and Passive Presentations: how to present your work, and how to get the most out of seminars, lectures and poster sessions.