## Bio

I’m a postdoc in the Stats Department at Oxford University, with Jonathan Marchini and Steve Smith. My research is in applications of machine learning to high dimensional data in neuroscience and genetics. I focus on modelling dependencies among phenotypes using transfer-learning, kernel methods, *LDS/HMMs*, relational models and Bayesian graphical models. Currently, I’m trying to uncover heritabiliy in brain networks (I’m looking at data from UK BIOBANK and the Human Connectome Project).

On the theory side, I’m interested in large GWAS studies for brain phenotypes. I’m also interested in *FDR* for *GWAS*, population structure+admixture (work with Maria De Iorio), L_1-regularization/selection, architectures for deep learning, dependent measures and hierarchical Dirichlet processes, mixture models & nonparametric mixtures, and pseudo-marginal methods in *MCMC/PMCMC/SMC*. I’m now a member of the common room at Kellogg College, where I act as an advisor. My *PhD* work was with Yee Whye Teh at the Gatsby Unit, where I developed non-parametric Bayesian models for population genetics.

I used to work at a *DNA* sequencing company: Halcyon Molecular, and I do some consulting work. I made the internet art piece mondrian-shapes and also the game Rogue, Fighter, Mage (a 7DRL). I wrote the back end for ‘Jennifer Lyn Morone™ Inc’, an art project covered by The Guardian (Human for sale: the artist who turned herself into a corporation) and The Economist (The incorporated woman).

## Publications §

- S. Schwab, R. Harbord, V. Zerbi,
__L.T. Elliott__, S. Afyouni, J.Q. Smith, M.W. Woolrich, S.M. Smith and T.E. Nichols. Directed functional connectivity using dynamic graphical models. 2017. Submitted to NeuroImage. __L.T. Elliott__, K. Sharp, F. Alfaro-Almagro, G. Douaud, K. Miller, J. Marchini, S. Smith. The genetic basis of human brain structure and function: 1,262 genome-wide associations found from 3,144 GWAS of multimodal brain imaging phenotypes from 9,707 UK Biobank participants. 2017. biorxiv preprint 2017/08/21/178806.- C. Bycroft, C. Freeman, D. Petkova, G. Band,
__L.T. Elliott__, K. Sharp, A. Motyer, D. Vukcevic, O. Delaneau, J. O'Connell, A. Cortes, S. Welsh, G. McVean, S. Leslie, P. Donnelly, J. Marchini. Genome-wide genetic data on ~500,000 UK Biobank participants. 2017. biorxiv preprint 2017/07/20/166298. __L.T. Elliott__, Y.W. Teh.*A nonparametric HMM for genetic imputation and coalescent inference*. 2016. Electronic Journal of Statistics. 10(2). p3425-51.*Code*.__L.T. Elliott__.*Bayesian nonparametric models of genetic variation*. 2016. PhD Thesis, University College London.- M. De Iorio,
__L.T. Elliott__, S. Favaro, K. Adhikari, Y.W. Teh.*Modeling population structure under hierarchical Dirichlet processes*. 2015. ArXiv preprint 1503.08278.*Code*. __L.T. Elliott__, Y.W. Teh.*Scalable imputation of genetic data with a discrete fragmentation-coagulation process*. 2012. Proceedings of the 26th Conference on Neural Information Processing Systems.- R.M. Cichy, P. Sterzer, J. Heinzle,
__L.T. Elliott__, F. Ramirez, J.D. Haynes.*Probing principles of large-scale object representation: Category preference and location encoding*. 2012. Human Brain Mapping. 34(7). p1636-51. - I. Murray,
__L.T. Elliott__.*Driving Markov chain Monte Carlo with a dependent random stream*. 2012. ArXiv preprint 1204.3187.*Code*. - Y.W. Teh, C. Blundell,
__L.T. Elliott__.*Modelling genetic variations using fragmentation-coagulation Processes*. 2011. Proceedings of the 25th Conference on Neural Information Processing Systems. - Y. Chen, P. Namburi,
__L.T. Elliott__, J. Heinzle, C. Soon, M. Chee, J.D. Haynes.*Cortical surface-based searchlight decoding*. 2010. Neuroimage. 56(2). p582-92.

## Slides

- The beta process: survival analysis, latent feature models, and the Indian buffet process. Lecture for the course
*Bayesian Nonparametrics*at the Gatsby unit, Spring 2012. - Expectation propagation. Tutorial for the course
*Unsupervised Learning*at the Gatsby unit, Summer 2011.

## Tools and Tricks

I’ve written a variety of tools to aid in statistical analyses (with a focus on genetic data). Some of these tools are available here on my homepage. I’ve also listed some useful code snippets.

- T. Fast transpose for comma, tab and whitespace separated variable files, with a small memory footprint. (Code in
*perl*.) - sgamfit. Fit a shifted gamma distribution. (Code in
*MATLAB*.) - gene2snps. Find all SNPs in a given gene (according to reference genome GRCh38 and SNP database snp146). (Code in
*MATLAB*.) - gene2range. Find the chromosome location of a gene (according to reference genome hg19). (Code in
*MATLAB*.) - crp_check. Test whether or not a set of clusterings are drawn from a Dirichlet process. (Code in
*scala*.) - GeneralizedStirlingS1. Compute a generalized Stirling number of the first kind. (Code in
*mathematica*.) This function is denoted S^{-1,-d}_{N,\ k} and is defined to be the coefficient of \xi^N inN!\ /\ k!\cdot\left(\sum_{j=1}^\infty(-d\ )^{\ j-1}\cdot\Gamma\left(\ j-1/d\ \right)\ /\ j!\cdot\xi^j\ /\ \Gamma\left(\ 1-1/d\ \right)\right)^k.I wrote this because I couldn’t find it in any computer algebra system, so I hope it helps. `lad`

. (*octave*code.) Rescale one vector so that it approximates another, with a L_1 penalty. This*octave*code snippet minimizes \sum_{i=1}^n |\ f_i - a g_i\ |, for the real number a \in \mathbb{R}, given the vectors f,g\in \mathbb{R}^n. This function uses the*octave*linear programming package`glpk`

.function a = lad(f, g) n = size(f, 1); c = [ones(n, 1); 0]; A = [eye(n) -g ; eye(n) g]; b = [-f; f]; x =

`glpk`

(c, A, b, [], [], repmat('L', 1, 2*n)); a = x(n + 1);

## Collaborations

Jonathan Marchini, Steve Smith, Andy Dahl, Victoria Hore, Kevin Sharp, Winnie Kretzschmar, Yee Whye Teh, Stefano Favaro, Maria De Iorio, Derek Aguiar and Barbara Engelhardt. Oxford BigBayes, Swhere Inc.

## Links

- LibBi. Probabilistic programming with SMC, optimized for parallel code. (A probabilistic programming language by Lawrence Murray).
- HomomorphicEncryption. Privacy in computation. Allows you to process data in an encrypted space, so only the data provider sees the result. (An \text{R} package by Louis Aslett.)
- Guess the Correlation. How good are you at guessing correlation coefficients from scatter plots? Test your skills! (A game by Omar Wagih.)
- pqR. Pretty quick \text{R}. An implementation of the core of \text{R} which improves the speed substantially. It’s inside a fork (\text{R-}2.15.0 or above), so all extended commands and third party packages still work. (Software by Radford Neal.)
- f.lux. Slowly redshift your monitor as the evening goes on. There’s some scientific evidence that this helps with sleep. (An app by Lorna and Michael Herf.)
- 7 Day Roguelike Challenge. Create a roguelike in 7 days. Entries are judged on creativity, completeness, aesthetics, scope and playability. (A competition by the 7DRL Committee.) My 2017 entry is Rogue, Fighter, Mage.

## Contact info

t: +（44）7456 ᐧ 060 ᐧ 153

e: elliott ᘒ stats ᐧ ox ᐧ ac ᐧ uk

a: Department of Statistics,
University of Oxford,
24\text{-}29 St Giles’,
OX1 ᐧ 3LB, Oxford, UK

s: LinkedIn ᐧ github ᐧ Google Scholar ᐧ hackernews ᐧ slashdot ᐧ |
| ᐧ facebook |