Interests
My core research interests are in the theory and methodology of statistical learning and inference, and, in particular, in penalized and pseudo-likelihood theory and methods, statistical computing and algorithms for regression problems and methods for clustering.
I also engage in interdisciplinary applied work of the kind that involves a real synthesis of approaches and has the potential to generate new advances in statistical learning and inference with broader impact. The applied work I am/have been involved in is in: sports science (modelling of high-frequency in-game events in team sports, and uncovering the links between human behaviour, health, fitness and overall well-being); finance (modelling the dynamics of financial indicators with structural dependencies); earthquake engineering (assessment of the vulnerability of the built environment from post-hazard survey data); neuroimaging (regression methods for brain lesions from MRI data and the summarization and visualization of effects); and genetics (infering changes in genomic network structures).
Last but not least, I am particularly interested in the design and development of scientific (and not only) software, to deliver the advances from my theoretical/methodological efforts and some tools that I find useful to the Data Science community and more broadly.
Research groups and themes
Some groups and themes that I founded or participate/have participated are:
Turing Interest Group on Data Science for Sports, Activity, and Well-being (founding member, organizer; 2017-2022)
Turing Interest Group on Machine Learning by Systems, for Systems (member; 2019-2020)
Turing Interest Group on Online Machine Learning (member; 2017-2020)
Statistics in Sports and Health research group (founder and leader; 2015-2017)
Statistics for Health Economic Evaluation research group (member; 2015-2017)
EPICentre research group (member; 2014-2017)
General Theory and Methodology and Computational Statistics research themes at the Department of Statistical Science, UCL (member; 2010-2017)
Preprints and unpublished reports
Kosmidis I, Zietkiewicz P (2023). Jeffreys-prior penalty for high-dimensional logistic regression: A conjecture about aggregate bias.
ArXiV Supplementary material Theory MethodsSterzinger P, Kosmidis I (2023). Diaconis-Ylvisaker prior penalized likelihood for \(p /n \to \kappa \in (0, 1)\) logistic regression.
ArXiV Supplementary material Theory MethodsZietkiewicz P, Kosmidis I (2023). Bounded-memory adjusted scores estimation in generalized linear models with large data sets.
ArXiV Supplementary material Theory MethodsKindalova P, Veldsman M, Nichols T E, Kosmidis I (2021). Penalized generalized estimating equations for relative risk regression with applications to brain lesion data.
bioRxiv Methods ApplicationsKöll S, Kosmidis I, Kleiber C, Zeileis A (2021). Bias reduction as a remedy to the consequences of infinite estimates in poisson and tobit regression.
ArXiV Theory MethodsFirth D, Kosmidis I, Turner H L (2019). Davidson-Luce model for multi-item choice with ties.
ArXiV TheoryKarimalis E, Kosmidis I, Peters G W (2017). Multi yield curve stress-testing framework incorporating temporal and cross tenor structural dependencies
SSRN Bank of England Staff Working Paper Series Methods ApplicationsKosmidis I, Passfield L (2015). Linking the performance of endurance runners to training and physiological effects via multi-resolution elastic net.
ArXiV Applications MethodsKosmidis I, Karlis D (2010). Supervised sampling for clustering large data sets.
CRiSM Working Paper Series Applications MethodsKosmidis I (2010). On iterative adjustment of responses for the reduction of bias in binary regression models.
CRiSM Working Paper Series Methods Theory
Publications
Wang Z, Dellaportas P, Kosmidis I (2023+). Bayesian Tensor Factorisations for Time Series of Counts.
The 15th Asian Conference on Machine Learning (ACML 2023) (one of the 27 accepted Journal Track Papers)
Machine Learning (accepted)
ArXiV ACML Theory MethodsKosmidis I, Lunardon N (2023+). Empirical bias-reducing adjustments to estimating functions.
Journal of the Royal Statistical Society: Series B (to appear)
DOI ArXiV Supplementary material Theory MethodsKosmidis I (2023). Mean and median bias reduction: A concise review and application to adjacent-categories logit models.
In: Kateri, M., Moustaki, I. (eds) Trends and Challenges in Categorical Data Analysis. Statistics for Social and Behavioral Sciences. Springer, Cham.
DOI ArXiV Supplementary material Methods ApplicationsPanos, A, Kosmidis I, Dellaportas P (2023). Scalable Marked Point Processes for Exchangeable and Non-Exchangeable Event Sequences.
In: Proceedings of The 26th International Conference on Artificial Intelligence and Statistics (AISTATS 2023), PMLR 206:236-252
PMLR ArXiV Supplementary material Methods ApplicationsNarayanan S, Kosmidis I, Dellaportas P (2023+). Flexible marked spatio-temporal point processes with applications to event sequences from association football.
Discussion paper in the Journal of the Royal Statistical Society: Series C (to appear)
DOI ArXiV Methods Applications
DOI: D Karlis (Proposer) DOI: L Egidi (Seconder) DOI: M Stival, L Schiavon DOI: J Mateu DOI: P Smith DOI: R Yurko, R Nugent DOI: Authors’ replySterzinger P, Kosmidis I (2023). Maximum softly-penalized likelihood for mixed effects logistic regression.
Statisics and Computing, 33, 53
DOI ArXiV Supplementary material Methods ApplicationsBellio R, Kosmidis I, Salvan A, Sartori N (2023). Parametric bootstrap inference for stratified models with high-dimensional nuisance specifications.
Statistica Sinica, 33
DOI ArXiV Theory MethodsBartlett T E, Kosmidis I, Silva R (2021). Two-way sparsity for time-varying networks with applications in genomics.
Annals of Applied Statistics, 15, 856-879
DOI ArXiv Methods Applications
Kindalova P, Kosmidis I, Nichols T E (2021). Voxel-wise and spatial modelling of binary lesion masks: Comparison of methods with a realistic simulation framework.
NeuroImage, 236, 118090
DOI bioRxiv Methods ApplicationsKosmidis I, Firth D (2021). Jeffreys-prior penalty, finiteness and shrinkage in binomial-response generalized linear models.
Biometrika, 108, 71–82
DOI ArXiV Supplementary material Theory Methods
- Whitaker G A, Silva R, Edwards D, Kosmidis I (2021). A Bayesian inference approach for determining player abilities in football.
Journal of the Royal Statistical Society: Series C, 70, 174-201
DOI ArXiV Methods Applications
- Veldsman M, Kindalova P, Husain M, Kosmidis I, Nichols T E (2020). Spatial distribution and cognitive impact of cerebrovascular risk-related white matter hyperintensities.
NeuroImage: Clinical, 28, 102405
DOI bioRxiv Methods Applications
- Turner H L, van Etten J, Firth D, Kosmidis I (2020). Modelling rankings in R: the PlackettLuce package.
Computational Statistics, 35, 1027–1057
DOI ArXiV Software Methods
- Kosmidis I, Kenne Pagui E C, Sartori N (2020). Mean and median bias reduction in generalized linear models.
Statistics and Computing, 30, 43-59
DOI ArXiV Methods Theory
Di Caterina C, Kosmidis I (2019). Location-adjusted Wald statistic for scalar parameters.
Computational Statistics and Data Analysis, 138, 126-142
DOI ArXiv Methods TheoryKyriakou S, Kosmidis I, Sartori N (2019). Median bias reduction in random-effects meta-analysis and meta-regression.
Statistical Methods in Medical Research, 28, 1622-1636
DOI ArXiV Supplementary material Methods ApplicationsTsokos A, Narayanan S, Kosmidis I, Baio G, Cucuringu M, Whitaker G, Király F J (2019). Modeling outcomes of soccer matches.
Machine Learning, 108, 77-95
DOI ArXiV ApplicationsIoannou I, Bessason B, Kosmidis I, Bjarnason J Ö, Rossetto T (2018). Empirical seismic vulnerability assessment of Icelandic buildings affected by the 2000 sequence of earthquakes.
Bulletin of Earthquake Engineering, 16, 5875–5903
DOI ApplicationsFrick H, Kosmidis I (2017). trackeR: Infrastructure for running and cycling data from GPS-enabled tracking devices in R.
Journal of Statistical Software, 82
DOI Software Methods ApplicationsKosmidis I, Guolo A, Varin C (2017). Improving the accuracy of likelihood-based inference in meta-analysis and meta-regression.
Biometrika, 104, 489-496
DOI ArXiV Theory Methods ApplicationsKosmidis I, Karlis D (2016). Model-based clustering using copulas with applications.
Statistics and Computing, 26, 1079–1099
DOI ArXiV Methods ApplicationsMaqsood T, Edwards M, Ioannou I, Kosmidis I, Rossetto T, Corby N (2016). Seismic vulnerability functions for Australian buildings by using GEM empirical vulnerability assessment guidelines.
Natural Hazards, 80, 1625-1650
DOI ApplicationsPanayi E, Peters G W, Kosmidis I (2015). Liquidity commonality does not imply liquidity resilience commonality: A functional characterisation for ultra-high frequency cross-sectional LOB data.
Quantitative Finance, 15, 1737-1758
DOI ArXiV ApplicationsAmes M, Peters G W, Bagnarosa G, Kosmidis I (2015). Upside and downside risk exposures of currency carry trades via tail dependence.
In: Glau, K, Scherer, M and Zagst, R (Eds.), Innovations in Quantitative Risk Management, Volume 99 of Springer Proceedings in Mathematics Statistics, 163-181
DOI ArXiV Applications MethodsKosmidis I (2014). Bias in parametric estimation: reduction and useful side-effects.
WIRE Computational Statistics, 6, 185-196
DOI ArXiV MethodsKosmidis I (2014). Improved estimation in cumulative link models.
Journal of the Royal Statistical Society: Series B, 76, 169-196
DOI ArXiV Supplementary Theory MethodsGrün B, Kosmidis I, Zeileis A (2012). Extended Beta regression in R: Shaken, stirred, mixed, and partitioned.
Journal of Statistical Software, 48
DOI Software MethodsKosmidis I, Firth D (2011). Multinomial logit bias reduction via the Poisson log-linear model.
Biometrika, 98, 755-759
DOI Theory MethodsLatuszynski K, Kosmidis I, Papaspiliopoulos O, Roberts G O (2011). Simulating events of unknown probabilities via reverse time martingales.
Random Structures and Algorithms, 38 , 441-452
DOI MethodsKosmidis I, Firth D (2010). A generic algorithm for reducing bias in parametric estimation.
Electronic Journal of Statistics, 4 1097-1112
DOI R Code and an example Methods TheoryKosmidis I, Firth D (2009). Bias reduction in exponential family nonlinear models.
Biometrika, 96, 793-804
DOI TheoryKosmidis I (2008). The profileModel R package: Profiling objectives for models with linear predictors.
R News, R Foundation for Statistical Computing, 8/2, 12-18.
Link Software Methods
PhD thesis
- Kosmidis I (2007). Bias reduction in exponential family nonlinear models (errata)
Selected presentations
- Package and collaboration networks in CRAN. Oxford R User Group, Oxford, UK, November 2018
ikosmidis_cranly_OxfordRUG_2018.R has the R code used in the presentation (note that outputs may vary as CRAN changes) - Location-adjusted Wald statistics. Institute for Statistics and Mathematics, WU Wien, Vienna, Austria, May 2018
- Reduced-bias estimation for models with ordinal responses. CEN ISBS 2017 Joint Conference, Vienna, Austria, August 2017
- Reduced-bias inference for multi-dimensional Rasch models with applications. 28th International Workshop on Statistical Modelling, Palermo, Italy, July 2013
- Bias reduction in generalized nonlinear models. Joint Statistical Meetings 2009, Washington, DC, 2009
- Profiling the parameters of models with linear predictors. useR 2008, Dortmund, Germany, August 2008