My Interests

Network Data

Networks are abundant in day to day life, from social networks to transportation. The connected nature of networks leads to statistical challenges: the classic assumption of data independence no longer holds. Classically, networks with only on/off connections have been considered.
My research in this area has focused on a Stochastic Block Model where connections a more general. I have considered two main problems: determining the number of groups in a network and detecting changes in group membership.

MCMC

In Bayesian statistics, lots of problems involving computing an integral. For lots of interesting models, these integrals cannot be solved by hand, but can be approximated by sampling from the model and computing averages.
In complex models, even sampling can be difficult. Markov chain Monte Carlo algorithms are a class of procedures that allow one to draw samples from any model.
I am interested in developing novel and more efficient MCMC algorithms.

Tuning MCMC

Most Markov chain Monte Carlo algorithms involve some parameters that the user is free to choose. These are often critical to the performance of the procedure, with poor chooses leading to higher variance estimates.
I am interested in theoretical and methodological ways to automate the choice of parameters for Hamiltonian Monte Carlo algorithms to alleviate this burden of choice.

Software

Practitioners in statistics rely heavily on software for data analysis. Modern research in statistics, especially methodology, produces a lot of code to demonstrate its worth.
Reproducability of research involving software is a big issue which I have been exploring, involving writing legible, interpretable code; teaching others on good practises in software; and version control.

My Research

  • Software

  • R package: SBMSplitMerge

    Github CRAN

    Ludkin, M (2020).

    A companion package to the paper "Inference for the stochastic block model with unknown number of blocks and non-conjugate edge models" arXiv


  • Academic Papers

  • Hug and Hop: a discrete-time, non-reversible Markov chain Monte Carlo algorithm.

    arXiv link

    Ludkin, M., Sherlock, C. (2019). arXiv pre-print

  • Inference for the stochastic block model with unknown number of blocks and non-conjugate edge models.

    Journal link DOI

    Matthew Ludkin, (2020). Computational Statistics & Data Analysis.

    arXiv link

    Ludkin, M (2020). arXiv pre-print

  • Dynamic stochastic block models: Parameter estimation and detection of changes in community structure.

    Journal link

    Ludkin, M., Eckley, I. and Neal, P. (2018). Statistics and Computing, 28, 1201-1213.


  • PhD Thesis

  • The autoregressive stochastic block model with changes in structure

    STOR-i, Lancaster University - 2014 - 2018

    A PDF version of my thesis is available via the Lancaster research portal


  • Funding Awards

  • e-COST short term scientific mission

    2017

    This award provided for a two week research trip to Paris, France for continued collaboration with Sté phane Robin (agrosParisTech) and Catherine Matias (UPMC).

  • STOR-i Research Grant

    2016

    This award provided funding to establish a research project titled "Fast Approximate Inference for Changes in Network Structure" in collaboration with Stéphane Robin (agrosParisTech) and Catherine Matias (UPMC).

  • Presentations Talks Posters

  • Talk: Hug 'n' Hop MCMC

    Reading - October 2018

    Contributed talk at An afternoon of Bayesian Computation. Slides available soon

  • Poster: Hug 'n' Hop MCMC

    Edinburgh - June 2018

    Contributed poster at ISBA world meeting conference. Poster available soon

  • Talk: Dynamic stochastic block models: parameter estimation and detection of changes in community structure

    Lancaster - January 2018

    Invited alumni speaker for the STOR-i annual conference 2018.

  • Poster: Detecting the number of communities in a network with general edge weights

    Lancaster - November 2017

    Emergent and Self Adaptive Systems workshop, DSI, Lancaster University.

  • Poster: Changes in Network Structure

    Cambridge - November 2016

    Challenges in Industry and Society Open for Business Event at Isaac Newton Institute, Cambridge.

  • Talk: Reversible Jump MCMC for changes in network structure

    Lancaster - April 2016

    Analysis of nonstationary multivariate time series Workshop at Lancaster University.

  • Talk: Reversible Jump MCMC for changes in network structure

    Lyon, France - April 2016

    Invited seminar speaker at Laboratoire de Biometrie et Biologie Evolutive at the University Claude Bernard, Lyon, France

  • Talk: Reversible Jump MCMC for changes in network structure

    Paris, France - April 2016

    Invited seminar speaker at INRA/MIA at AgrosParisTech, Paris, France

  • Education

  • MRes

    2013 - 2014

    Distinction - Lancaster University

  • MSci Mathematics

    2009 - 2013

    1st Class with Honours - University of Birmingham

  • Prize for Sciences

    2011

    University of Birmingham

  • Prize for Statistics

    2010

    University of Birmingham

  • The Alan Murray Centenary Scholarship

    2010-2013

    University of Birmingham

    I feel very privileged to have been granted this scholarship, which helped me greatly in my studies during my time at Birmingham.