BEGIN:VCALENDAR
VERSION:2.0
PRODID:researchseminars.org
CALSCALE:GREGORIAN
X-WR-CALNAME:researchseminars.org
BEGIN:VEVENT
SUMMARY:Matthew Lee (University of Bristol)
DTSTART:20201015T130000Z
DTEND:20201015T140000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/1
DESCRIPTION:Title: EpiViz: an implementation of Circos plots for epidemiolog
ists\nby Matthew Lee (University of Bristol) as part of (ED-3S) Essex
Data Science Seminar Series\n\n\nAbstract\nEpidemiology studies predominan
tly focus on single exposure and single outcome associations. However\, bi
ological pathways involve numerous processes and identifying meaningful in
termediate associations that can be taken forward for further analysis is
complex. This is particularly the case for studies involving metabolomics
data\, as effects rarely occur in isolation. Gaining global overview of hu
ndreds of exposure/outcome associations may therefore aid downstream analy
ses. Visual inspection is one of the main modes of understanding global ex
posure/outcome associations. EpiViz is a wrapper that makes producing Cric
os plots simple and efficient for those new to programming and data visual
isation.\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/1/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Godwin Osuntoki (University of Essex)
DTSTART:20201022T130000Z
DTEND:20201022T140000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/2
DESCRIPTION:Title: Bayesian Analysis of chromosomal interactions in Hi-C dat
a using the hidden Markov random field model\nby Godwin Osuntoki (Univ
ersity of Essex) as part of (ED-3S) Essex Data Science Seminar Series\n\n\
nAbstract\nThere are different biological methods that have been developed
over the years for analysis of the 3D structure of the DNA. Few computati
onal and statistical methods have\, however\, been developed to analysis d
ata generated using the Hi-C method. We follow statistical methodology to
explore the Hi-C data. The Hi-C data is well suited to be analyzed using a
finite mixture model. The Potts model\, a hidden Markov random field mode
l\, was employed to analyze the hidden (latent) components. The hidden com
ponents through the Potts model can be categorized into k components (k =
2\,3…\,K). Using the Metropolis-within-Gibbs approach to analyze the dat
a\, the proposed method was able to detect interactions (short and long ra
nge) and loops. A large part of the significant interactions that we detec
t are found within Topological Associated Domains\, which is one of the 3D
structures known to occur in Hi-C data.\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/2/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Nosheen Faiz (University of Essex)
DTSTART:20201105T140000Z
DTEND:20201105T150000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/4
DESCRIPTION:Title: Assessing how feature selection and hyper-parameters infl
uence optimal trees ensemble and random projection\nby Nosheen Faiz (U
niversity of Essex) as part of (ED-3S) Essex Data Science Seminar Series\n
\n\nAbstract\nOur work investigates the effect of feature selection on thr
ee methods: Random Forest (Breiman 2001)\, Optimal Trees Ensemble (Khan et
al 2016) and Random Projection (Canning and Samworth 2017) in high dimens
ional settings. To this end\, LASSO has been considered for selecting the
most important features based on training data for dimension reduction. Ad
ditionally\, the influence of various hyper-parameters regulating the thre
e methods has also been assessed. Analysis on several benchmark datasets i
s given to illustrate the phenomena. The results reveal that feature selec
tion improves the predictive performance of the Random Forest and Random P
rojection methods in addition to reducing the computational burden. The pe
rformance of Optimal Trees Ensemble is less influenced by feature selectio
n.\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/4/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Peng Liu (University of Essex)
DTSTART:20201112T140000Z
DTEND:20201112T150000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/5
DESCRIPTION:Title: Ordering and Inequalities for Mixtures on Risk Aggregatio
n\nby Peng Liu (University of Essex) as part of (ED-3S) Essex Data Sci
ence Seminar Series\n\n\nAbstract\nAggregation sets\, which represent mode
l uncertainty due to unknown dependence\, are an important object in the s
tudy of robust risk aggregation. In this talk\, we investigate ordering re
lations between two aggregation sets for which the sets of marginals are r
elated by two simple operations: distribution mixtures and quantile mixtur
es. Intuitively\, these operations ``homogenize" marginal distributions
by making them similar. As a general conclusion from our results\, more ``
homogeneous" marginals lead to a larger aggregation set\, and thus more se
vere model uncertainty\, although the situation for quantile mixtures is m
uch more complicated than that for distribution mixtures. \nWe proceed t
o study inequalities on the worst-case values of risk measures in risk agg
regation\, which represent conservative calculation of regulatory capital.
Among other results\, we obtain an order relation on VaR under quantile
mixture for marginal distributions with monotone densities. Numerical resu
lts are presented to visualize the theoretical results and further inspire
some conjectures.\nFinally\, we discuss the connection of our results to
joint mixability and to merging p-values in multiple hypothesis testing.\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/5/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Tolulope Fadina (University of Essex)
DTSTART:20210225T140000Z
DTEND:20210225T150000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/6
DESCRIPTION:Title: Symmetric measures of variability induced by risk measure
s\nby Tolulope Fadina (University of Essex) as part of (ED-3S) Essex D
ata Science Seminar Series\n\n\nAbstract\nGeneral measures of variability
induced by risk measures are investigated for their potential applications
to risk management. We emphasize on the three classes of variability meas
ures generated by the Value-at-Risk\, Expected Shortfall\, and the Expecti
les. Their properties are explored\, and we obtain a characterization resu
lt on general model spaces. Convergence properties and asymptotic normalit
y of the empirical variability measures estimators are established. An app
lication of the variability measures to financial data is also investigate
d.\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/6/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Ioana Olan (University of Cambridge)
DTSTART:20201126T140000Z
DTEND:20201126T150000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/7
DESCRIPTION:Title: Detecting the hierarchical structure of the cell nucleus<
/a>\nby Ioana Olan (University of Cambridge) as part of (ED-3S) Essex Data
Science Seminar Series\n\n\nAbstract\nChromatin consists of DNA wrapped a
round histones and forms complex three-dimensional structures within the c
ell nucleus with various degrees of compaction. Genes have been shown to b
e repressed by their proximity to the nuclear periphery or activated by be
ing in contact with special regulatory regions called enhancers. Thus the
relative positioning of genes and their interactions with other regions ar
e very important in determining whether they are expressed or not. Interac
tions between pairs of genomic regions have been studied using assays such
as Hi-C\, which generate large matrices estimating interaction frequencie
s. We use such interaction estimates as weights in a network whose nodes a
re equally sized genomic regions and perform nested community detection in
order to resolve the relative positioning of genomic regions of interest
and model the interior of the cell nucleus. Our biological model is cellul
ar senescence\, a phenotype associated with dramatic changes in its chroma
tin interactions network relative to normal cells. Senescence corresponds
to permanent cell cycle arrest and has been shown to act as a protective b
arrier against tumourigenesis.\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/7/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Josh Bull (University of Oxford)
DTSTART:20201203T140000Z
DTEND:20201203T150000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/8
DESCRIPTION:Title: Can maths tell us how to win at Fantasy Football?\nby
Josh Bull (University of Oxford) as part of (ED-3S) Essex Data Science Se
minar Series\n\n\nAbstract\nFantasy Football is an online game played by m
illions of people every year\, in which players attempt to predict the out
come of football matches over the course of a season. To the surprise of e
veryone (including myself)\, I was lucky enough to be crowned the winner o
f the 2019-20 Fantasy Premier League\, one of the largest competitions in
the UK. As a researcher in Mathematical Oncology at the University of Oxfo
rd\, people have asked me whether I used maths to win – while I followed
some strategies at the time\, I didn’t have any proof that they were in
some sense mathematically optimal. However\, mathematical modelling is a
tool which is capable of exploring exactly these kinds of questions: how c
an we identify the best strategies to tackle complex problems? What types
of data are important to consider\, and how should we use them to inform o
ur decisions? In this talk\, I’ll analyse how different quantitative app
roaches can be used to tackle key questions in Fantasy Football\, and iden
tify the strengths and weaknesses of these frameworks. Finally\, I’ll ad
dress the question: Can maths tell us how to win at Fantasy Football?\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/8/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Osama Mahmoud (University of Essex)
DTSTART:20210211T140000Z
DTEND:20210211T150000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/9
DESCRIPTION:Title: Slope-Hunter: A robust method for index-event bias correc
tion in genome-wide association studies of conditional analyses\nby Os
ama Mahmoud (University of Essex) as part of (ED-3S) Essex Data Science Se
minar Series\n\n\nAbstract\nBackground: Studying genetic associations with
prognosis (e.g. survival\, subsequent events) is problematic due to selec
tion bias - also termed index event bias or collider bias - whereby select
ion on disease status can induce associations between causes of incidence
with prognosis. A current method for adjusting genetic associations for th
is bias assumes there is no genetic correlation between incidence and prog
nosis\, which may not be a plausible assumption.\n\nMethods: We propose an
alternative\, the ‘Slope-Hunter’ approach\, which is unbiased even wh
en there is genetic correlation between incidence and prognosis. Our appro
ach has two stages. First\, we use cluster-based techniques to identify: v
ariants affecting neither incidence nor prognosis (these should not suffer
bias and only a random sub-sample of them are retained in the analysis)\;
variants affecting prognosis only (excluded from the analysis). Second\,
we fit a cluster-based model to identify the class of variants only affect
ing incidence\, and use this class to estimate the adjustment factor. {\\c
olor{blue} The underlying assumption of our approach is that variants affe
cting only incidence explain more variation in incidence than any group of
variants with unique effects\, e.g. via same exposure\, on both incidence
and prognosis}.\n\nResults: Simulation studies showed that {\\color{blue}
our approach eliminates the bias and outperforms alternatives in the pres
ence of genetic correlation\, and performs as well as alternatives under n
o genetic correlation when its assumption is satisfied. We applied the ‘
Slope-Hunter’ method to a study of fasting blood insulin levels (FI) con
ditional on body mass index (BMI)\, estimated the index event bias\, and a
djusted conditional associations of the lead variants with FI. Our estimat
es suggested that there were common causes of BMI and FI of concordant dir
ections of effect\, that are in-line with previously observed association
between obesity and insulin resistance.}\n\nConclusions: Our approach is u
nbiased even in the presence of genetic correlation between incidence and
progression when the underlying assumptions hold. Bias-adjusting methods s
hould be used to carry out causal analyses when conditioning on incidence.
\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/9/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Yanchun Bao (University of Essex)
DTSTART:20201217T140000Z
DTEND:20201217T150000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/10
DESCRIPTION:Title: Estimating mode effects from a sequential mixed-modes ex
periment\nby Yanchun Bao (University of Essex) as part of (ED-3S) Esse
x Data Science Seminar Series\n\n\nAbstract\nThe large-scale household pan
el study Understanding Society (The U.K. Household Longitudinal Study UKHL
S) has\, until recently\, used interviewers to administer its questionnair
es\, but is now in the process of allowing individuals to participate usin
g the web. Survey data are known to be affected by survey mode so a sequen
tial mode-effects experiment was carried out on to evaluate the impact of
this change on the panel. In this talk we present a novel estimator and an
alysis strategy to quantify the impact of mode across a wide range of vari
ables\, with large mode effects on the covariance of a pair of variables u
sed to indicate an increased risk that statistical analyses involving this
pair will be affected.\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/10/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Rafal Kulakowski (University of Essex)
DTSTART:20210204T140000Z
DTEND:20210204T150000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/11
DESCRIPTION:by Rafal Kulakowski (University of Essex) as part of (ED-3S) E
ssex Data Science Seminar Series\n\nAbstract: TBA\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/11/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Yassir Rabhi (University of Essex)
DTSTART:20201210T140000Z
DTEND:20201210T150000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/12
DESCRIPTION:Title: Copulas and measures of dependence under length-biased s
ampling and informative censoring\nby Yassir Rabhi (University of Esse
x) as part of (ED-3S) Essex Data Science Seminar Series\n\n\nAbstract\nLen
gth-biased data are often encountered in cross-sectional surveys and preva
lent-cohort studies on disease durations. Under length-biased sampling sub
jects with longer disease durations have greater chance to be observed. As
a result\, covariate values linked to the longer survivors are favoured b
y the sampling mechanism. When the sampled durations are also subject to r
ight censoring\, the censoring is informative. Modelling dependence struct
ure without adjusting for these issues leads to biased results. In this ta
lk\, I will present a study on copulas for modelling dependence when the c
ollected data are length-biased and account for both informative censoring
and covariate bias. I will address the nonparametric estimation of the bi
variate distribution\, copula function and its density\, and Kendall and S
pearman measures for right-censored length-biased data. The proposed estim
ator of the bivariate CDF is a Hadamard-differentiable functional of two M
LEs\, Kaplan-Meier and empirical CDF\, and inherits their efficiencies. Ba
sed on this estimator\, we devise estimators for copula function and a loc
al-polynomial estimator for copula density that accounts for boundary bias
. In addition\, I will introduce estimators for Kendall and Spearman measu
res. The weak convergence of the estimators will also be discussed. The pr
oposed method is then applied to analyse a set of right-censored length-bi
ased data on survival with dementia\, collected as part of a nationwide st
udy in Canada.\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/12/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Carolin Strobl (Universität Zürich)
DTSTART:20201119T140000Z
DTEND:20201119T150000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/13
DESCRIPTION:Title: A Statistician’s Botanical Garden - The Ideas behind T
rees\, Model-Based Trees and Random Forests\nby Carolin Strobl (Univer
sität Zürich) as part of (ED-3S) Essex Data Science Seminar Series\n\n\n
Abstract\nClassification and regression trees\, model-based trees and rand
om forests are powerful statistical methods from the field of machine lear
ning. They have been shown to achieve a high prediction accuracy\, especia
lly in big data applications with many predictor variables and complex ass
ociation patterns (such as nonlinear and higher-order interaction effects)
. While individual trees are easy to interpret\, random forests are "black
box" prediction methods. They do\, however\, provide variable importance
measures\, that are being used to judge the relevance of the individual pr
edictor variables. The aim of this presentation is to introduce the ration
ale behind trees\, model-based trees and random forests\, to illustrate th
eir potential for high-dimensional data exploration\, e.g.\, in psychologi
cal research\, but also to point out limitations and potential pitfalls in
their practical application.\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/13/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Shenggang Hu (University of Essex)
DTSTART:20221013T130000Z
DTEND:20221013T140000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/14
DESCRIPTION:Title: Statistical disaggregation - a Monte Carlo approach for
imputation under constraints\nby Shenggang Hu (University of Essex) as
part of (ED-3S) Essex Data Science Seminar Series\n\nLecture held in NTC.
1.04.\n\nAbstract\nStatistical disaggregation has become more and more imp
ortant for smart energy systems. A typical example of such disaggregation
problems is to learn energy consumption for a higher resolution level (dat
a recorded at higher frequency) based on data at a lower resolution (data
recorded at lower frequency). Constrained models are often used in such pr
oblems and they are often very useful compared to their unconstrained coun
terparts in terms of reducing uncertainty and leading to an improvement of
the overall performance. However\, these constrained models usually are n
ot expressible as ordinary distributions due to their intractable density
functions which makes it hard to conduct further analysis. This paper intr
oduces a novel constrained Monte Carlo sampling algorithm based on Langevi
n diffusions and rejection sampling to solve the problem of sampling from
constrained models. This new method is then applied to a statistical disag
gregation problem for an electricity consumption dataset. Our approach pr
ovides excellent accuracy of data imputation\, based on our simulation stu
dies and data analysis. The new method is also justified theoretically.\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/14/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Prof Christian Martin Hennig (University of Bologna\, UCL)
DTSTART:20221103T140000Z
DTEND:20221103T150000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/15
DESCRIPTION:Title: Advances in using cluster analysis for species delimitat
ion\nby Prof Christian Martin Hennig (University of Bologna\, UCL) as
part of (ED-3S) Essex Data Science Seminar Series\n\nLecture held in STEM
3.1.\n\nAbstract\nBiological species are often delimited based on genetic
multilocus data using methods for inferring phylogenetic trees or model- o
r distance-based cluster analysis. A major problem here is that genetic di
ssimilarity does not only arise from separated species\, but also if subpo
pulations of a species live in geographically distant areas without geneti
c exchange. In any case\, be it using partitioning cluster analysis or hie
rarchical trees\, it is a hard problem to decide the number of species\, a
nd whether groups that are candidates for being species actually belong to
gether. I will discuss some the use of some new approaches for clustering
and estimating the number of clusters for this problem\, focusing particul
arly on testing whether observed genetic heterogeneity within a species ca
ndidate group can be explained be geographical distance rather than consis
ting of separate species. This requires hypothesis testing in a distance-d
istance regression model. I will also discuss the integration of such a te
sting routine in a fully automated method for species delimitation.\n\nRef
erence\n\nHausdorf\, B\, Hennig\, C. Species delimitation and geography. M
ol Ecol Resour. 2020\; 20: 950– 960. https://doi.org/10.1111/1755-0998.1
3184\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/15/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Dr Johan van der Molen (University of Cambridge)
DTSTART:20221124T140000Z
DTEND:20221124T150000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/16
DESCRIPTION:Title: Dirichlet process mixture inconsistency for the number o
f components: how worried should we be in practice?\nby Dr Johan van d
er Molen (University of Cambridge) as part of (ED-3S) Essex Data Science S
eminar Series\n\nLecture held in STEM 3.1.\n\nAbstract\nBayesian nonparame
tric mixture models are widely used for model-based clustering due to thei
r flexibility and conceptual simplicity\, as well as the availability of
efficient sampling methods for performing inference. However\, recent work
has established that such models have undesirable asymptotic properties r
egarding the estimation of the number of clusters. For instance\, Dirichle
t Process Mixtures (DPMs) have been shown to be inconsistent for the numbe
r of clusters\, and overestimation of the number of clusters has been obse
rved in practice for finite samples. Finite mixtures with a prior on the n
umber of components - also known as Mixtures of Finite Mixtures (MFMs) - h
ave been suggested as an asymptotically consistent alternative\, but the e
ffects of model misspecification can still result in asymptomatic inconsis
tency and poor estimation of the number of clusters in practice. \n\nHere
we specifically focus on estimation of the number of clusters in Bayesian
nonparametric mixtures in practice\, including the impact of Markov chain
Monte Carlo (MCMC) post-processing algorithms for summarisation and identi
fication of a final representative summary clustering. We consider practic
al scenarios of low to moderate dimension\, through both simulation studie
s and applications to real biomolecular data. In the situations we conside
r\, we confirm that even when the parametric form of the mixture component
distributions is correctly specified\, DPMs lead to mild overestimation o
f the number of clusters for finite samples. However\, we also demonstrate
that this can be corrected by common summarisation methods\, suggesting t
hat applications of DPMs in practice may be more robust than the theory mi
ght suggest. We show that\, for both DPMs and MFMs\, mixture component den
sity misspecification typically leads to more dramatic overestimation\, wi
th DPMs providing slightly worse estimates than MFMs\, but with the common
pattern of “true” clusters in the data being split into smaller subcl
usters due to additional mixture components being required to flexibly cap
ture features of the data inadequately described by the misspecified model
s. We consider implications for high-dimensional data analysis\, in which
simplifying assumptions that are commonly made in practice for computation
al tractability (e.g. assuming a diagonal covariance matrix for Gaussian m
ixture components) are also expected to result in model misspecification.
As part of our work\, we compare popular MCMC post-processing algorithms f
or identifying a final summary clustering\, and show that although some of
them have a positive impact on results\, others can introduce severe over
estimation of the number of clusters\, even when the underlying posterior
distribution from which samples are being drawn is centred on the true num
ber of clusters. This is joint work with Yannis Chaumeny\, Paul Kirk\, Ant
hony Davidson.\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/16/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Dr Alexei Vernitski (University of Essex)
DTSTART:20221027T130000Z
DTEND:20221027T140000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/17
DESCRIPTION:Title: Using machine learning to solve mathematical problems an
d to search for examples and counterexamples in pure maths research\nb
y Dr Alexei Vernitski (University of Essex) as part of (ED-3S) Essex Data
Science Seminar Series\n\nLecture held in STEM 3.1.\n\nAbstract\nOur recen
t research can be generally described as applying state-of-the-art technol
ogies of machine learning to suitable mathematical problems. We use both r
einforcement learning and supervised learning (underpinned by deep learnin
g). As to mathematical problems we consider\, they include learning to unt
angle a braid (this problem is not unlike the problem of solving the Rubik
cube)\, learning to find the parity of a permutation (as compared to the
classical problem of deep learning of learning to find the parity bit of a
binary array)\, comparing mathematical mistakes made by artificial intell
igence with those made by human mathematicians\, etc.\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/17/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Qiuyi Hong (University of Essex)
DTSTART:20221117T140000Z
DTEND:20221117T150000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/18
DESCRIPTION:Title: A Bilevel Game-TheoreDc Decision-Making Framework for St
rategic Retailers in Both Local and Wholesale Electricity Markets\nby
Qiuyi Hong (University of Essex) as part of (ED-3S) Essex Data Science Sem
inar Series\n\nLecture held in STEM 3.1.\n\nAbstract\nIn this talk we prop
ose a bilevel game-theoretic model for multiple strategic retailers partic
ipating in both wholesale and local electricity markets while considering
customers’ switching behaviours. At the upper level\, each retailer maxi
mizes its own profit by making optimal offering decisions in the retail ma
rket and bidding decisions in the day-ahead wholesale (DAW) and local powe
r exchange (LPE) markets. The interaction among multiple strategic retaile
rs is formulated using the Bertrand competition model. For the lower level
\, there are three optimisation problems. First\, the customers’ welfare
maximisation problem with their switching behaviors is formulated to capt
ure the demand responses from customers. Second\, a market-clearing proble
m is formulated for the independent system operator (ISO) in the DAW marke
t. Third\, a novel LPE market is developed for retailers to facilitate the
ir power balancing. In addition\, the bilevel multi-leader multi-follower
Stackelberg game forms an equilibrium problem with equilibrium constraints
(EPEC) problem\, which is solved by the diagonalization algorithm. Numeri
cal results demonstrate the feasibility and effectiveness of the EPEC mode
l and the importance of modeling customers’ switching behaviors. We corr
oborate that incentivising customers’ switching behaviors and increasing
the number of retailers facilitates retail competition\, which results in
reducing strategic retailers’ retail prices and profits. Moreover\, the
relationship between customers’ switching behaviors and welfare is refl
ected by a balance between the electricity purchasing cost (i.e.\, electri
city price) and the electricity consumption level.\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/18/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Mateo Salles (University of Essex)
DTSTART:20230209T140000Z
DTEND:20230209T150000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/19
DESCRIPTION:Title: Supervised Learning for Untangling Braids\nby Mateo
Salles (University of Essex) as part of (ED-3S) Essex Data Science Seminar
Series\n\nLecture held in STEM 3.1.\n\nAbstract\nUntangling a braid is a
typical multi-step process\, and reinforcement learning can be used to tra
in an agent to untangle braids. Here we present another approach. Starting
from the untangled braid\, we produce a dataset of braids using breadth-f
irst search and then apply behavioral cloning to train an agent on the out
put of this search. As a result\, the (inverses of) steps predicted by the
agent turn out to be an unexpectedly good method of untangling braids\, i
ncluding those braids which did not feature in the dataset.\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/19/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Dr. Peng Liu (University of Kent)
DTSTART:20230504T130000Z
DTEND:20230504T140000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/20
DESCRIPTION:Title: Optimal Smooth Approximation for Quantile Matrix Factori
sation\nby Dr. Peng Liu (University of Kent) as part of (ED-3S) Essex
Data Science Seminar Series\n\nLecture held in STEM 3.1.\n\nAbstract\nMatr
ix Factorisation (MF) is essential to many estimation tasks. Most existing
matrix factorisation methods focus on least squares matrix factorisation
(LSMF)\, which aims to minimise a smooth L2 loss between observations and
their dependent matrix measurement variables. In reality\, however\, L1 lo
ss and check loss are widely used in regression to deal with outliers or o
bservations contaminated by skewed or heavy-tailed noise. Although under c
ertain conditions\, linear convergence to the global optimality can be est
ablished for matrix factorisation under the L2 loss\, there is a lack of p
rovably efficient algorithms for solving matrix factorisation under non-sm
ooth losses. In this paper\, we investigate Quantile Matrix Factorization
(QMF)\, the counterpart of Quantile Regression in matrix estimation\, that
adopts a tunable check loss and introduces robustness to matrix estimatio
n for skewed and heavy tailed observations\, which are prevalent in realit
y. To deal with the non-smooth loss\, we propose Nesterov smoothed QMF (Ns
QMF)\, extending Nesterov’s optimal smooth approximation technique to th
e matrix factorisation setting. We then present an alternating minimizatio
n algorithm to solve the smooth NsQMF efficiently. We mathematically prove
that solving the smoothed NsQMF is equivalent to solving the original non
-smooth QMF problem and that our proposed algorithm achieves linear conver
gence to the global optimality of QMF. Numerical evaluations verify our th
eoretical findings and demonstrate that NsQMF significantly outperforms th
e commonly used LSMF and prior approximate smoothing heuristics for QMF un
der various noise distributions.\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/20/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Dr. Xiaochuan Yang (University of Brunel)
DTSTART:20230525T130000Z
DTEND:20230525T140000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/21
DESCRIPTION:Title: Some recent progress in random geometric graphs: beyond
the standard regimes\nby Dr. Xiaochuan Yang (University of Brunel) as
part of (ED-3S) Essex Data Science Seminar Series\n\nLecture held in STEM
3.1.\n\nAbstract\nI will survey some recent joint works with Mathew Penros
e (Bath) on the cluster structure of random geometric graphs in a regime
that is less discussed in the literature. The statistics of interest incl
ude the number of k-components\, the number of components\, the number of
vertices in the giant component\, and the connectivity threshold. We show
LLN and normal/Poisson approximation by Stein's method.\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/21/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Dr. Yufei Zhang (London School of Economics & Political Science)
DTSTART:20230511T130000Z
DTEND:20230511T140000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/22
DESCRIPTION:Title: Exploration-exploitation trade-off for continuous-time r
einforcement learning\nby Dr. Yufei Zhang (London School of Economics
& Political Science) as part of (ED-3S) Essex Data Science Seminar Series\
n\nLecture held in STEM 3.1.\n\nAbstract\nRecently\, reinforcement learnin
g (RL) has attracted substantial research interests. Much of the attention
and success\, however\, has been for the discrete-time setting. Continuou
s-time RL\, despite its natural analytical connection to stochastic contro
ls\, has been largely unexplored and with limited progress. In particular\
, characterising sample efficiency for continuous-time RL algorithms remai
ns a challenging and open problem.\n\nIn this talk\, we develop a framewor
k to analyse model-based reinforcement learning in the episodic setting. W
e then apply it to optimise exploration-exploitation trade-off for linear-
convex RL problems\, and report sublinear (or even logarithmic) regret bou
nds for a class of learning algorithms inspired by filtering theory. The a
pproach is probabilistic\, involving analysing learning efficiency using c
oncentration inequalities for correlated continuous-time observations\, an
d applying stochastic control theory to quantify the performance gap betwe
en applying greedy policies derived from estimated and true models.\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/22/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Prof. Chenggui Yuan (Swansea University)
DTSTART:20230601T130000Z
DTEND:20230601T140000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/24
DESCRIPTION:Title: Numerical solutions of SDEs with irregular coefficients<
/a>\nby Prof. Chenggui Yuan (Swansea University) as part of (ED-3S) Essex
Data Science Seminar Series\n\nLecture held in STEM 3.1.\n\nAbstract\nStoc
hastic differential equations (SDEs) with irregular coefficients have been
widely studied. In this talk\, I will discuss the strong convergence and
the weak convergence of SDEs with irregular coefficients. The convergenc
e rate will be investigated under different irregular conditions on coeffi
cients.\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/24/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Dr. Robert Gaunt (The University of Manchester)
DTSTART:20230615T130000Z
DTEND:20230615T140000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/25
DESCRIPTION:Title: Normal approximation for the posterior in exponential fa
milies\nby Dr. Robert Gaunt (The University of Manchester) as part of
(ED-3S) Essex Data Science Seminar Series\n\nLecture held in STEM 3.1.\n\n
Abstract\nIn this talk I'll introduce quantitative Bernstein-von Mises typ
e bounds on the normal approximation of the posterior distribution in expo
nential family models when centering either around the posterior mode or a
round the maximum likelihood estimator. Our bounds\, obtained through a ve
rsion of Stein’s method\, are non-asymptotic\, and data dependent\; they
are of the correct order both in the total variation and Wasserstein dist
ances\, as well as for approximations for expectations of smooth functions
of the posterior. All our results are valid for univariate and multivaria
te posteriors alike\, and do not require a conjugate prior setting. We ill
ustrate our findings on a variety of exponential family distributions\, in
cluding Poisson\, multinomial and normal distribution with unknown mean an
d variance. The resulting bounds have an explicit dependence on the prior
distribution and on sufficient statistics of the data from the sample\, an
d thus provide insight into how these factors may affect the quality of th
e normal approximation.\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/25/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Dr. Arthur Maheo (Amazon)
DTSTART:20230622T130000Z
DTEND:20230622T140000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/26
DESCRIPTION:Title: Benders decomposition for public transportation\nby
Dr. Arthur Maheo (Amazon) as part of (ED-3S) Essex Data Science Seminar Se
ries\n\nLecture held in STEM 3.1.\n\nAbstract\nCanberra (Australia) wants
to design a transportation network combining high-frequency buses with on-
demand taxis. The resulting hub-and-shuttle network design problem is a la
rge\, difficult mixed-integer program. We identified how to decompose the
problem – design first\, route second – and used a modern Benders deco
mposition on the resulting formulation.\nThis new approach is orders of ma
gnitude faster\, allowing us to solve full instances where a standard appr
oach can only do small ones.\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/26/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Prof Boris Mirkin (National Research University Higher School of E
conomics)
DTSTART:20231006T120000Z
DTEND:20231006T130000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/27
DESCRIPTION:Title: Anomalous clustering at various data formats\nby Pro
f Boris Mirkin (National Research University Higher School of Economics) a
s part of (ED-3S) Essex Data Science Seminar Series\n\nLecture held in 1N1
.4.1.\n\nAbstract\nAnomalous clustering is a method for extracting cluster
s one-by-one. It is an extension of the Principal Component Analysis meth
od to zero-one matrix factorization settings. After a brief overview of va
rious versions of the method\, including its extensions to similarity da
ta\, spatial data\, and fuzzy clustering\, I am going to concentrate on a
most recent development\, a triple-stage application of the approach to t
he analysis of spatial-temporal patterns in a coastal oceanic phenomenon o
f upwelling (see Nascimento et al. 2023).\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/27/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Dr Jin Zhu (LSE)
DTSTART:20231019T130000Z
DTEND:20231019T140000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/28
DESCRIPTION:Title: A Tuning-Free Algorithm for Sparsity-Constraint Optimiza
tion\nby Dr Jin Zhu (LSE) as part of (ED-3S) Essex Data Science Semina
r Series\n\nLecture held in STEM 3.1.\n\nAbstract\nSparsity-constraint opt
imization has wide applicability in signal processing\, statistics\, and m
achine learning. Existing fast algorithms must burdensomely tune parameter
s\, such as the step size or the implementation of precise stop criteria\,
which may be challenging to determine in practice. To address this issue\
, we develop an algorithm named sparsity-constraint optimization via splic
ing iteration (SCOPE) to optimize nonlinear differential objective functio
ns with strong convexity and smoothness in low dimensional subspaces. Algo
rithmically\, the SCOPE algorithm converges effectively without tuning par
ameters. Theoretically\, SCOPE has a linear convergence rate and converges
to a solution that recovers the true support set when it correctly specif
ies the sparsity. We also develop parallel theoretical results without res
tricted-isometry-property-type conditions. We apply SCOPE’s versatility
and power to solve sparse quadratic optimization\, learn sparse classifier
s\, and recover sparse Markov networks for binary variables. The numerical
results on these specific tasks reveal that SCOPE perfectly identifies th
e true support set with a 10–1000 speedup over the standard exact solver
\, confirming SCOPE’s algorithmic and theoretical merits. Our open-sourc
e Python package scope based on C++ implementation is publicly available o
n GitHub\, reaching a ten-fold speedup on the competing convex relaxation
methods implemented by the cvxpy library.\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/28/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Dr Shenggang Hu (University of Warwick)
DTSTART:20231026T130000Z
DTEND:20231026T140000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/29
DESCRIPTION:Title: Differential Privacy of Bayesian Posterior under Contami
nation\nby Dr Shenggang Hu (University of Warwick) as part of (ED-3S)
Essex Data Science Seminar Series\n\nLecture held in STEM 3.1.\n\nAbstract
\nIn recent years\, differential privacy has been adopted by tech-companie
s and governmental agencies as the standard for measuring privacy in algor
ithms. We study the level of differential privacy in Bayesian posterior sa
mpling setups. As opposed to the common privatization approach of injectin
g Laplace/Gaussian noise into the output\, Huber's contamination model is
considered\, where we replace at random the data points with samples from
a heavy-tailed distribution. The derived bound for the differential privac
y level in our approach matches the existing literature while lifting the
restriction on bounded observation space. We further consider the effect o
f sample size on privacy level and conclude that asymptotically the contam
ination approach is fully private at no cost of information loss.\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/29/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Prof Wolfgang Hardle (Humboldt-Universität zu Berlin\, Germany)
DTSTART:20240118T140000Z
DTEND:20240118T150000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/30
DESCRIPTION:Title: Data Science in a Math-Less Digital Society\nby Prof
Wolfgang Hardle (Humboldt-Universität zu Berlin\, Germany) as part of (E
D-3S) Essex Data Science Seminar Series\n\nLecture held in STEM 3.1.\n\nAb
stract\nIn an increasingly digital and data-driven world\, the importance
of data science cannot be overstated. Data science\, by itself\, carries
a "push to analyse“ button though\, that lets the analyst forget about
the „math behind the machine learning tools“\n\nWe cover a few example
s\, where data science needs math in order to be understood and applied.\n
\nBy the end of this talk\, attendees will gain a fresh perspective on dat
a science's role in a math-less digital society. They will leave with prac
tical insights\, tools\, and strategies to leverage data effectively\, fos
tering a culture of data-driven decision-making that transcends mathematic
al barriers.\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/30/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Dr Dimitra Kosta (University of Edinburgh)
DTSTART:20231123T134500Z
DTEND:20231123T144500Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/31
DESCRIPTION:Title: Maximum likelihood estimation of toric Fano varieties\nby Dr Dimitra Kosta (University of Edinburgh) as part of (ED-3S) Essex
Data Science Seminar Series\n\nLecture held in Zoom.\n\nAbstract\nI will t
alk about the maximum likelihood estimation problem for several classes of
toric Fano models. I will start by exploring the maximum likelihood degre
e for all 2-dimensional Gorenstein toric Fano varieties. I will show that
the ML degree is equal to the degree of the surface in every case except f
or the quintic del Pezzo surface with two ordinary double points and provi
de explicit expressions that allow one to compute the maximum likelihood e
stimate in closed form whenever the ML degree is less than 5. I will explo
re the reasons for the ML degree drop using A-discriminants and intersecti
on theory. If there is time\, I will discuss about toric Fano varieties as
sociated to 3-valent phylogenetic trees and their ML degree.\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/31/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Dr Richard Mann (University of Leeds)
DTSTART:20240201T140000Z
DTEND:20240201T150000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/33
DESCRIPTION:Title: Collective decision-making by rational agents\nby Dr
Richard Mann (University of Leeds) as part of (ED-3S) Essex Data Science
Seminar Series\n\nLecture held in STEM 3.1.\n\nAbstract\nThe decisions mad
e by others are a valuable source of social information about the world\,
because they may have knowledge that we lack. This means that when one age
nt makes a given choice\, it can induce others to do so as well. In this t
alk I will describe a theory of rational agents who optimally utilise the
social information provided by others\, and explore the dynamics this prod
uces at the individual and group level. In particular\, I will show how th
e implicit beliefs such agents hold about the physical and social environm
ent shape their response to each other\, and how changes to the environmen
t that conflict with these beliefs can dramatically alter collective behav
iour and impact the success of groups.\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/33/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Dr Jinyu Tian (Macau University of Science and Technology)
DTSTART:20231214T140000Z
DTEND:20231214T150000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/34
DESCRIPTION:Title: Discreteness Problem in Adversarial Machine Learning
\nby Dr Jinyu Tian (Macau University of Science and Technology) as part of
(ED-3S) Essex Data Science Seminar Series\n\n\nAbstract\nAdversarial exam
ples (AEs) of deep neural networks (DNNs) are receiving ever-increasing at
tention because they help in understanding the mechanism of DNNs and provi
de a novel perspective of the ethics of deep learning applications. In man
y real scenarios\, AEs have to be discrete (e.g. digital images). Most exi
sting works achieve the discreteness relying on the discretization of cont
inuous AEs. Unfortunately\, they cannot sufficiently control the spatial d
ifference before and after discretizing continuous AEs\, which will leads
to two sid-effects: degrading the attack capability of the obtained discre
te AEs or introducing the extra distortion. \n\nIn this work\, we propose
an adversarial attack called Discrete Attack (DATK) to produce continuous
AEs tightly close to their discrete counterparts. Owning the negligible sp
atial distance between them\, the expected discrete AEs perform with the s
ame powerful attack capability as the continuous AEs without an extra dist
ortion overhead. More precisely\, the proposed DATK generate AEs from a no
vel perspective by directly modeling adversarial perturbations (APs) as di
screte random variables. The AE generation problem thus reduces to the est
imation of the distribution of discrete APs. Since this problem typically
is nondifferential\, we relax it with the proposed reparameterizing tricks
and obtain an approximated continuous distribution of discrete APs. Our t
heoretical proof shows that\, by virtue the continuous APs sampled from th
e approximated distribution\, the spatial distance between the resultant c
ontinuous AEs and their discrete counterparts are tightly bounded\, which
significantly overcomes the side-effects caused by the discretization. Ext
ensive results over Imagenet\, Cifar10 and TU Berlin Sketch demonstrate th
e superiority of our method when attacking representative DNNs including V
gg19\, Resnet50\, DenseNet121 and MobilenetV2. It is also verified that ou
r DATK is more robust against the state-ofthe-art adversarial detection me
thods.\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/34/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Dr Hong Duong (University of Birmingham)
DTSTART:20231130T140000Z
DTEND:20231130T150000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/35
DESCRIPTION:Title: Model Reduction of Complex Systems\nby Dr Hong Duong
(University of Birmingham) as part of (ED-3S) Essex Data Science Seminar
Series\n\nLecture held in STEM 3.1.\n\nAbstract\nComplex systems in nature
and in applications (such as molecular systems\, crowd dynamics\, swarmin
g\, opinion formation\, just to name a few) are often described by systems
of stochastic differential equations (SDEs) and partial differential equa
tions (PDEs). It is often analytically impossible or computationally prohi
bitively expensive to deal with the full models due to their high dimensio
nality (degrees of freedom\, number of involved parameters\, etc.). It is
thus of great importance to approximate such large and complex systems by
simpler and lower dimensional ones\, while still preserving the essential
information from the original model. This procedure is referred to as mode
l reduction or coarse-graining in the literature. In this talk\, I will pr
esent methods for qualitative and quantitative coarse-graining of several
SDEs and PDEs\, in the presence or absence of a scale-separation.\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/35/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Dr Yuyu Chen (University of Melbourne)
DTSTART:20231116T130000Z
DTEND:20231116T140000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/36
DESCRIPTION:Title: Diversification of infinite-mean Pareto distributions\nby Dr Yuyu Chen (University of Melbourne) as part of (ED-3S) Essex Data
Science Seminar Series\n\nLecture held in Zoom.\n\nAbstract\nWe show the
perhaps surprising inequality that the weighted average of negatively depe
ndent super-Pareto random variables\, possibly caused by triggering events
\, is larger than one such random variable in the sense of first-order sto
chastic dominance. The class of super-Pareto distributions is extremely he
avy-tailed and it includes the class of infinite-mean Pareto distributions
. We discuss several implications of this result via an equilibrium analys
is in a risk exchange market. First\, diversification of super-Pareto loss
es increases portfolio risk\, and thus a diversification penalty exists. S
econd\, agents with super-Pareto losses will not share risks in a market e
quilibrium. Third\, transferring losses from agents bearing super-Pareto l
osses to external parties without any losses may arrive at an equilibrium
which benefits every party involved. The empirical studies show that our n
ew inequality can be observed empirically for real datasets that fit well
with extremely heavy tails.\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/36/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Dr Xiaochun Meng (University of Bath)
DTSTART:20240509T130000Z
DTEND:20240509T140000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/37
DESCRIPTION:Title: Angular Combining of Forecasts of Probability Distributi
ons\nby Dr Xiaochun Meng (University of Bath) as part of (ED-3S) Essex
Data Science Seminar Series\n\nLecture held in STEM 3.1.\n\nAbstract\nWhe
n multiple forecasts are available for a probability distribution\, foreca
st combining enables a pragmatic synthesis of the information to extract t
he wisdom of the crowd. A linear opinion pool has been widely used\, where
by the combining is applied to the probability predictions of the distribu
tional forecasts. However\, it has been argued that this will tend to deli
ver overdispersed distributional forecasts\, prompting the combination to
be applied\, instead\, to the quantile predictions of the distributional f
orecasts. Results from different applications are mixed\, leaving it as an
empirical question whether to combine probabilities or quantiles. In this
paper\, we present an alternative approach. Looking at the distributional
forecasts\, combining the probability forecasts can be viewed as vertical
combining\, with quantile forecast combining seen as horizontal combining
. Our proposal is to allow combining to take place on an angle between the
extreme cases of vertical and horizontal combining. We term this angular
combining. The angle is a parameter that can be optimized using a proper s
coring rule. For implementation\, we provide a pragmatic numerical approac
h and a simulation algorithm. Among our theoretical results\, we show that
\, as with vertical and horizontal averaging\, angular averaging results i
n a distribution with mean equal to the average of the means of the distri
butions that are being combined. We also show that angular averaging produ
ces a distribution with lower variance than vertical averaging\, and\, und
er certain assumptions\, greater variance than horizontal averaging. We pr
ovide empirical support for angular combining using weekly distributional
forecasts of Covid mortality.\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/37/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Mahendra Singh Rajpoot (University of Essex)
DTSTART:20240125T140000Z
DTEND:20240125T150000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/38
DESCRIPTION:Title: Large Language Models: A Stepping Stone for AGI!\nby
Mahendra Singh Rajpoot (University of Essex) as part of (ED-3S) Essex Dat
a Science Seminar Series\n\nLecture held in STEM 3.1.\n\nAbstract\nIn the
rapidly evolving landscape of Artificial Intelligence (AI)\, Large Languag
e Models (LLMs) have emerged as a transformative force\, showcasing remark
able capabilities in natural language understanding and generation. This p
resentation delves into the pivotal role that LLMs play as a stepping ston
e towards achieving Artificial General Intelligence (AGI). We explore the
fundamental principles\, applications\, and underlying mechanisms that pro
pel LLMs while contemplating their implications for the broader goal of AG
I. The talk will navigate through recent advancements\, challenges\, and e
thical considerations in harnessing the potential of LLMs\, ultimately env
isioning their contribution to the evolution of comprehensive artificial i
ntelligence\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/38/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Dr Yi Zhang (University of Birmingham)
DTSTART:20240425T130000Z
DTEND:20240425T140000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/40
DESCRIPTION:Title: On discounted Markov decision processes and their extens
ions\nby Dr Yi Zhang (University of Birmingham) as part of (ED-3S) Ess
ex Data Science Seminar Series\n\nLecture held in 4SW.6.28.\n\nAbstract\nT
he theory for discounted Markov decision processes (MDPs) has been well de
veloped. In this talk we review some basic results concerning their occupa
tion measures\, which are convenient for the studies of optimal control pr
oblems with constraints. After that\, we discuss the possibility of their
extensions to more general models (uniformly absorbing MDPs\, absorbing MD
Ps\, or more general MDPs with total criteria). The studies of absorbing M
DPs have been active recently.\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/40/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Dr Kareemah Chopra (University of Essex)
DTSTART:20240208T140000Z
DTEND:20240208T150000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/41
DESCRIPTION:Title: [Cancelled] The Bunching Behaviour of Cows\nby Dr Ka
reemah Chopra (University of Essex) as part of (ED-3S) Essex Data Science
Seminar Series\n\nLecture held in STEM 3.1.\n\nAbstract\nBunching behavior
in cattle may occur for several reasons including enabling social interac
tions\, a response to stress or danger\, or due to shared interest in reso
urces such as feeding or watering areas. There is evidence in pasture graz
ed cattle that bunching may occur more frequently at higher ambient temper
atures\, possibly due to sharing of fly-load or to seek shade from the dir
ect sun under heat stress conditions. Here we demonstrate how bunching beh
avior is associated with higher ambient temperatures in a barn-housed UK d
airy herd. A real-time local positioning system (RTLS) was used\, as part
of a precision livestock farming (PLF) approach\, to track the spatial pos
ition and activity of a commercial dairy herd (c100 cows) in a freestall b
arn continuously at high temporal resolution for 4 mo between August and N
ovember 2014. Bunching was determined using 4 different spatial measures d
etermined on an hourly basis: herd full and core range size\, mean herd in
ter-cow distance (ICD)\, and mean herd nearest neighbor distance (NND). Fo
r hourly mean ambient temperatures above 20°C\, the herd showed higher bu
nching behavior with increasing ambient temperature (i.e.\, reduced full a
nd core range size\, ICD\, and NND). Aggregated space-use intensity was fo
und to positively correlate with localized variations in temperature acros
s the barn (as measured by animal mounted sensors)\, but the level of corr
elation decreased at higher ambient barn temperatures. Bunching behavior m
ay increase localized temperatures experienced by individuals and hence ma
y be a maladaptive behavioral response in housed dairy cattle\, which are
known to suffer heat stress at higher temperatures. Our study is the first
to use high-resolution positional data to provide evidence of association
s between bunching behavior and higher ambient temperatures for a barn-hou
sed dairy herd in a temperate region (UK). Further studies are needed to e
xplore the exact mechanisms for this response to inform both welfare and p
roduction management.\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/41/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Professor Richard J. Samworth (University of Cambridge)
DTSTART:20240229T140000Z
DTEND:20240229T150000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/42
DESCRIPTION:Title: Isotonic subgroup selection\nby Professor Richard J.
Samworth (University of Cambridge) as part of (ED-3S) Essex Data Science
Seminar Series\n\nLecture held in STEM 3.1.\n\nAbstract\nGiven a sample of
covariate-response pairs\, we consider the subgroup selection problem of
identifying a subset of the covariate domain where the regression function
exceeds a pre-determined threshold. We introduce a computationally-feasib
le approach for subgroup selection in the context of multivariate isotonic
regression based on martingale tests and multiple testing procedures for
logically-structured hypotheses. Our proposed procedure satisfies a non-as
ymptotic\, uniform Type I error rate guarantee with power that attains the
minimax optimal rate up to poly-logarithmic factors. Extensions cover cla
ssification\, isotonic\nquantile regression and heterogeneous treatment ef
fect settings.\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/42/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Professor Edward Rochead (Defence Science and Technology Laborator
y)
DTSTART:20240307T140000Z
DTEND:20240307T150000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/43
DESCRIPTION:Title: The Alliance for Data Science Professionals\nby Prof
essor Edward Rochead (Defence Science and Technology Laboratory) as part o
f (ED-3S) Essex Data Science Seminar Series\n\nLecture held in STEM 3.1.\n
\nAbstract\nThe talk will begin by introducing the Alliance\, its members
and how it was formed. It will then explain how individuals can become acc
redited as Advanced Data Science Professionals and also describe the plans
being formed to accredit degrees. It is expected that the discussion woul
d focus on how the AfDSP can work with academic colleagues and ensure accr
editation is attractive and meaningful to them\, and also consider how it
may feed into the employability of graduates in relevant disciplines.\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/43/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Laurel Ariane Regibeau-Rockett (Stanford University)
DTSTART:20240321T140000Z
DTEND:20240321T150000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/44
DESCRIPTION:Title: Hurricanes as heat engines\nby Laurel Ariane Regibea
u-Rockett (Stanford University) as part of (ED-3S) Essex Data Science Semi
nar Series\n\nLecture held in STEM 3.1.\n\nAbstract\nHurricanes are danger
ous and destructive atmospheric phenomena\, frequently causing loss of liv
es worldwide. Improving our understanding of hurricanes can help improve h
urricane forecasts and projections of their response to climate change. On
e conceptual model of the hurricane\, which has supported major advancemen
ts in hurricane science\, is the conceptualization of the hurricane as a h
eat engine. This theoretical framework supports research at the intersect
ion of physics\, mathematics\, and atmospheric science. In this seminar\,
we will review this important theoretical model and some of its applicatio
ns\, together with possible directions of future research in this interdis
ciplinary domain.\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/44/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Professor Mariachiara Di Cesare (University of Essex)
DTSTART:20240314T140000Z
DTEND:20240314T150000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/45
DESCRIPTION:Title: Institute of Public Health and Wellbeing opportunities t
o enhance research for all\nby Professor Mariachiara Di Cesare (Univer
sity of Essex) as part of (ED-3S) Essex Data Science Seminar Series\n\nLec
ture held in STEM 3.1.\n\nAbstract\nThe IPHW\, established in 2022\, repre
sents a major strategic innovation for the University of Essex\, bringing
together our community of experts to provide pioneering leadership in the
production of world-class research\, knowledge exchange and impact. Workin
g with regional\, national\, and international partners\, the IPHW is driv
en by a collective goal of creating a healthier and fairer society. During
this seminar we will discuss the IPHW mission\, vision\, and strategy and
look at opportunities to enhance interdisciplinary research in the field
of health and wellbeing with a special focus on data science.\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/45/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Dr Maria Brigida Ferraro (Sapienza University of Rome)
DTSTART:20240530T130000Z
DTEND:20240530T140000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/46
DESCRIPTION:Title: Two-mode clustering in a fuzzy setting: methods and clus
ter validity indices\nby Dr Maria Brigida Ferraro (Sapienza University
of Rome) as part of (ED-3S) Essex Data Science Seminar Series\n\nLecture
held in STEM 3.1.\n\nAbstract\nThe aim of clustering is to find a partitio
n of the rows (e.g. objects) of a data matrix based on the values assumed
on a set of variables (columns). Two objects belong to the same cluster if
the corresponding rows are close to each other according to a certain met
ric based on all the variables. However\, it can be reasonable to seek clu
sters such that objects assigned to the same cluster are close to each oth
er with respect to a subset of variables. The research\ninterest can also
be reversed\, i.e.\, the goal is to find clusters of variables close to ea
ch other in terms of a subset of objects. Standard clustering algorithms a
re not adequate to accomplish these tasks. For this purpose\, two-mode clu
stering methods have been introduced. Two-mode clustering consists in simu
ltaneously partitioning modes (e.g.\, objects and variables) of an observe
d two-mode data matrix.\n\nIn the literature\, two-mode clustering methods
have been extensively studied and extended\nalong various directions. Mos
t of them are based on the classical approach to clustering\, i.e.\, the o
bjects (or the variables) are either assigned or not to the clusters. A mo
re powerful and flexible exploratory approach is represented by introducin
g fuzziness in the clustering process. In this case\, the objects (or the
variables) are no longer either assigned or not to the clusters\, but belo
ng to the clusters with the so-called (fuzzy) membership degrees taking va
lues in the interval [0\,1]. A high membership degree\, close to 1\, recog
nizes an object (or variable) strongly assigned to a cluster\, i.e.\, an o
bject (or variable) very close to the corresponding cluster prototype.\n\n
Starting from the Double k-Means\, we propose a class of two-mode clusteri
ng algorithms in a\nfuzzy framework\, including some robust proposals\, ta
king into account that\, in this case\,\ndifferent kinds of outliers exist
and should be considered.\nIn addition\, in order to evaluate the two fuz
zy partitions and to choose the optimal numbers of clusters\, new cluster
validity indices are introduced. The proposed measures are defined in\nter
ms of the compactness within each cluster and separation between clusters.
Starting from\nsome well-known indices in standard fuzzy clustering\, som
e generalizations to the two-mode\ncase are addressed. The adequacy of the
proposals is checked by means of simulation and real-case studies.\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/46/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Dr Mohamed Bader (University of Portsmouth)
DTSTART:20240627T130000Z
DTEND:20240627T140000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/47
DESCRIPTION:Title: Beyond Theory: Machine Learning and AI in Action for Ear
ly Behavior and Outcome Prediction\nby Dr Mohamed Bader (University of
Portsmouth) as part of (ED-3S) Essex Data Science Seminar Series\n\nLectu
re held in 1N1.4.1.\n\nAbstract\nThis talk explores the practical applicat
ion of machine learning and AI in predicting early behaviors and outcomes
across healthcare\, digital marketing\, and industrial sectors. In healthc
are\, particularly ICUs\, we discuss how AI models forecast patient outcom
es and improve resource management. In digital marketing\, the focus is on
how AI anticipates consumer behaviors to optimize marketing strategies. L
astly\, in industrial applications\, we examine AI's role in predicting ma
intenance needs and enhancing operational reliability.\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/47/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Professor Guy Nason (Imperial College London)
DTSTART:20240516T130000Z
DTEND:20240516T140000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/48
DESCRIPTION:Title: Network Time Series\nby Professor Guy Nason (Imperia
l College London) as part of (ED-3S) Essex Data Science Seminar Series\n\n
Lecture held in STEM 3.1.\n\nAbstract\nA network time series is a multivar
iate time series where the individual series are known to be linked by som
e underlying network structure. Sometimes this network is known a priori\,
and sometimes the network has to be created\, often inferred from the mul
tivariate series itself. Network time series are becoming increasingly com
mon\, long\, and collected over a large number of variables. We are partic
ularly interested in network time series whose network structure changes o
ver time.\n\nWe describe some recent developments in the modeling of netwo
rk time series via generalized network autoregressive (GNAR) process model
s. These models use regular autoregressive links between a variable and it
s past and between a variable and the past of its neighbours. GNAR models
are highly parsimonious and\, hence\, work well for short series or those
afflicted by worrying amounts of missing data. For the same reason\, they
tend not to overfit and often exhibit excellent forecasting performance\,
especially when compared to alternatives such as vector autoregressive mod
els.\n\nThis talk explains the GNAR model and some interesting variants. W
e introduce some new tools for model selection and exhibit their use on ep
idemic and economic data.\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/48/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Dr Kareemah Chopra (University of Essex)
DTSTART:20240502T130000Z
DTEND:20240502T140000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/50
DESCRIPTION:Title: The Bunching Behaviour of Cows\nby Dr Kareemah Chopr
a (University of Essex) as part of (ED-3S) Essex Data Science Seminar Seri
es\n\n\nAbstract\nBunching behavior in cattle may occur for several reason
s including enabling social interactions\, a response to stress or danger\
, or due to shared interest in resources such as feeding or watering areas
. There is evidence in pasture grazed cattle that bunching may occur more
frequently at higher ambient temperatures\, possibly due to sharing of fly
-load or to seek shade from the direct sun under heat stress conditions. H
ere we demonstrate how bunching behavior is associated with higher ambient
temperatures in a barn-housed UK dairy herd. A real-time local positionin
g system (RTLS) was used\, as part of a precision livestock farming (PLF)
approach\, to track the spatial position and activity of a commercial dair
y herd (c100 cows) in a freestall barn continuously at high temporal resol
ution for 4 mo between August and November 2014. Bunching was determined u
sing 4 different spatial measures determined on an hourly basis: herd full
and core range size\, mean herd inter-cow distance (ICD)\, and mean herd
nearest neighbor distance (NND). For hourly mean ambient temperatures abov
e 20°C\, the herd showed higher bunching behavior with increasing ambient
temperature (i.e.\, reduced full and core range size\, ICD\, and NND). Ag
gregated space-use intensity was found to positively correlate with locali
zed variations in temperature across the barn (as measured by animal mount
ed sensors)\, but the level of correlation decreased at higher ambient bar
n temperatures. Bunching behavior may increase localized temperatures expe
rienced by individuals and hence may be a maladaptive behavioral response
in housed dairy cattle\, which are known to suffer heat stress at higher t
emperatures. Our study is the first to use high-resolution positional data
to provide evidence of associations between bunching behavior and higher
ambient temperatures for a barn-housed dairy herd in a temperate region (U
K). Further studies are needed to explore the exact mechanisms for this re
sponse to inform both welfare and production management.\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/50/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Anusa Suwanwong (University of Essex)
DTSTART:20240620T130000Z
DTEND:20240620T140000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/51
DESCRIPTION:Title: A Gene Selection Method for Classification with Three Cl
asses Using Proportional Overlapping Scores\nby Anusa Suwanwong (Unive
rsity of Essex) as part of (ED-3S) Essex Data Science Seminar Series\n\nLe
cture held in STEM 3.1.\nAbstract: TBA\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/51/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Dr Yu Wu (Southwest Jiaotong University\, China)
DTSTART:20241010T110000Z
DTEND:20241010T120000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/52
DESCRIPTION:Title: Efforts to overcome the curse of dimensionality in seque
ntial decision-making problems\nby Dr Yu Wu (Southwest Jiaotong Univer
sity\, China) as part of (ED-3S) Essex Data Science Seminar Series\n\n\nAb
stract\nSequential decision-making problems are widespread\, and solving t
hem is essential for enhancing efficiency\, reducing costs\, and optimizin
g resource allocation. However\, these problems are notoriously difficult
due to the “curse of dimensionality.” Given the diversity of sequentia
l decision-making problems and the broad applicability of solution methods
\, this\ntalk will primarily focus on the complex Dynamic Vehicle Routing
Problem (DVRP). It will start by elucidating the specific challenges posed
by the curse of dimensionality\, including the exponential growth of stat
e space\, action space\, and transition probabilities. Then\, the talk wil
l examine and discuss existing techniques to address these challenges\, su
ch as state aggregation\, initial policy generation\, offline-online polic
y improvement\, state (or state-action) value function representation\, an
d methods for updating and leveraging probabilistic laws. Finally\, by mod
ularly deconstructing\, updating\, and recombining these techniques\, this
talk will propose new approaches to potentially overcome the curse of dim
ensionality in the future.\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/52/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Prof Tengyao Wang (LSE)
DTSTART:20241128T120000Z
DTEND:20241128T130000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/53
DESCRIPTION:Title: High-dimensional changepoint estimation with heterogeneo
us missingness\nby Prof Tengyao Wang (LSE) as part of (ED-3S) Essex Da
ta Science Seminar Series\n\nLecture held in CTC 3.02.\n\nAbstract\nWe pro
pose a new method for changepoint estimation in partially-observed\, high-
dimensional time series that undergo a simultaneous change in mean in a sp
arse subset of coordinates. Our first methodological contribution is to i
ntroduce a 'MissCUSUM' transformation\, that captures the interaction betw
een the signal strength and the level of missingness in each coordinate.
In order to borrow strength across the coordinates\, we project these Miss
CUSUM statistics along a direction found as the solution to an optimisatio
n problem. The changepoint can then be estimated as the location of the p
eak of the projected series. In a model that allows different missingness
probabilities in different component series\, we identify that the key in
teraction between the missingness and the signal is an observation-probabi
lity-weighted sum of squares of the signal change in each coordinate. Mo
re specifically\, we prove that the angle between the estimated and oracle
projection directions\, as well as the changepoint location error\, are c
ontrolled with high probability by the sum of two terms\, both involving t
his weighted sum of squares\, and representing the error incurred due to n
oise and due to missingness respectively. The striking effectiveness of o
ur methodology is further demonstrated both on simulated data\, and on an
oceanographic data set.\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/53/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Dr Eleni-Rosalina Andrinopoulou (Erasmus Medical Center Rotterdam)
DTSTART:20241205T120000Z
DTEND:20241205T130000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/54
DESCRIPTION:Title: Assessing Risk Indicators in Clinical Practice with Join
t Models\nby Dr Eleni-Rosalina Andrinopoulou (Erasmus Medical Center R
otterdam) as part of (ED-3S) Essex Data Science Seminar Series\n\nLecture
held in https://essex-university.zoom.us/j/98548135065.\n\nAbstract\nThe i
ncreasing availability of clinical measures (e.g.\, electronic medical rec
ords) leads to collecting many different types of information. This inform
ation mainly includes multiple longitudinal measurements collected during
follow-up visits of the patient to the clinic. Big data is the key element
to new developments and precision medicine. Nowadays\, individualized\, d
ynamic predictions are popular in different medical fields because they im
prove patient care. In particular\, it is of high clinical interest to pre
dict future measurements to assess recovery on patients who experienced a
stroke or predict life expectancy for patients with Cystic Fibrosis.\n \nP
hysicians collect a variety of measurements over time to assess the severi
ty and progression of a disease. Even though all outcomes will be consider
ed together intuitively\, they will usually be analyzed separately. It is
biologically relevant to study them together\; therefore\, it is more appr
opriate to analyze them assuming a single statistical model. This\, howeve
r\, poses many challenges. In particular\, different characteristics of th
e patients' longitudinal profiles could influence the outcome(s) of intere
st. For example\, the rate of change could be a better predictor than the
actual value. Therefore\, it is essential to assume the correct associatio
n structure when obtaining dynamic predictions.\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/54/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Dr Victoria Volodina (University of Exeter)
DTSTART:20241114T120000Z
DTEND:20241114T130000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/55
DESCRIPTION:Title: Polynomial structural equation models (SEMs) for decisio
n support systems\nby Dr Victoria Volodina (University of Exeter) as p
art of (ED-3S) Essex Data Science Seminar Series\n\nLecture held in CTC 3.
02.\n\nAbstract\nProbabilistic graphical models are widely used to model c
omplex systems with uncertainty. In a decision-support context\, Bayesian
analysis may focus on obtaining beliefs from a decision maker and multiple
expert panels about individual variables or propagating the effects of ne
w evidence through the graph G\, coherently updating beliefs in vertices t
hat are not yet established. The associated computations can become cumber
some. To facilitate the decision-making process\, we propose adopting the
polynomial structural equation model (SEM) to depict complex relationships
between individual variables in graph G and with a utility function in po
lynomial form. Since the marginal posterior distributions of individual va
riables can become analytically intractable\, we develop a nonparametric m
essage-passing algorithm that propagates information throughout the graph
using only moments\, enabling exact calculation of expected utility scores
. We illustrate the proposed methodology with examples and an application
to decision problems in energy planning and healthcare.\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/55/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Dr Jonas Latz (Manchester University)
DTSTART:20250227T120000Z
DTEND:20250227T130000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/56
DESCRIPTION:Title: Cancelled\nby Dr Jonas Latz (Manchester University)
as part of (ED-3S) Essex Data Science Seminar Series\n\nLecture held in CT
C 3.02.\nAbstract: TBA\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/56/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Dr Jianya Lu (University of Essex)
DTSTART:20241017T110000Z
DTEND:20241017T120000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/57
DESCRIPTION:Title: Distribution estimation for time series via DNN-based GA
Ns with an application to change-point estimation\nby Dr Jianya Lu (Un
iversity of Essex) as part of (ED-3S) Essex Data Science Seminar Series\n\
nLecture held in CTC 3.02.\n\nAbstract\nThe generative adversarial network
s (GANs) have recently been applied to estimating the distribution of inde
pendent and identically distributed data\, and have attracted a lot of res
earch attention. In this talk\, I'll demonstrate the effectiveness of GAN
s in estimating the distribution of stationary time series. Theoretically\
, we derive a non-asymptotic error bound for the Deep Neural Network (DNN)
-based GANs estimator for the stationary distribution of the time series.
Our approach is based on the blocking technique and the $m$-dependence app
roximation technique that divides the time series into interlacing blocks
of equal size and then constructs independent blocks. Based on the theoret
ical analysis\, we propose an algorithm for estimating the change-point in
time series distribution. Numerical results of Monte Carlo experiments an
d real data application are given to validate our theory and algorithm.\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/57/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Porf. Michael J. Daniels (University of Florida\, USA)
DTSTART:20250213T120000Z
DTEND:20250213T130000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/58
DESCRIPTION:Title: A Bayesian nonparametric approach for evaluating the cau
sal effect of treatment in observational cohort studies with semi-competin
g risks\nby Porf. Michael J. Daniels (University of Florida\, USA) as
part of (ED-3S) Essex Data Science Seminar Series\n\nLecture held in CTC 3
.02.\n\nAbstract\nWe develop a Bayesian nonparametric (BNP) approach to ev
aluate the causal effect of treatment where a nonterminal event may be cen
sored by one (or more) terminal event(s)\, but not vice versa (i.e.\, semi
-competing risks). Based on the idea of principal stratification\, we defi
ne a novel estimand for the causal effect of treatment on the nonterminal
event. We introduce identification assumptions (using factorizations based
on vine copulas) indexed by sensitivity parameters and show how to draw
inference using our BNP approach. We illustrate our methodology using data
from a cardiovascular cohort study.\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/58/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Dr Soudeep Deb (Indian Institute of Management Bangalore)
DTSTART:20241212T120000Z
DTEND:20241212T130000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/59
DESCRIPTION:Title: Structural breaks in the spatial network of real estate
dynamics: A study of UK property transactions\nby Dr Soudeep Deb (Indi
an Institute of Management Bangalore) as part of (ED-3S) Essex Data Scienc
e Seminar Series\n\nLecture held in CTC 3.02.\n\nAbstract\nThe real estate
market is a dynamic system shaped by various economic\, social\, and envi
ronmental factors\, making the detection of structural changes crucial for
understanding market trends\, managing risks\, and guiding investment and
policy decisions. This study examines temporal changes in Greater London
’s real estate market using weekly data at the MSOA (Middle Layer Super
Output Areas) level through a two-stage methodology. First\, Local Indicat
ors of Spatial Association (LISA) are applied to identify significant clus
ters of high and low property prices and spatial outliers\, which are then
integrated into a network framework incorporating geographical distance.
Second\, structural breaks in market dynamics are detected using network L
aplacians\, capturing both gradual and abrupt shifts over time. These find
ings are further leveraged to develop a localized house price index and a
data-driven zoning approach\, offering enhanced tools for evaluating prope
rty prices with greater accuracy\, benefiting both investors and policymak
ers.\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/59/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Dr Wasiur R. KhudaBukhsh (University of Nottingham)
DTSTART:20250515T110000Z
DTEND:20250515T120000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/60
DESCRIPTION:Title: Enzyme kinetic reactions as interacting particle systems
: Stochastic averaging and parameter inference\nby Dr Wasiur R. KhudaB
ukhsh (University of Nottingham) as part of (ED-3S) Essex Data Science Sem
inar Series\n\nLecture held in CTC 3.02.\n\nAbstract\nWe consider a stocha
stic model of multistage Michaelis--Menten (MM) type enzyme kinetic reacti
ons describing the conversion of substrate molecules to a product through
several intermediate species. The high-dimensional\, multiscale nature of
these reaction networks presents significant computational challenges\, es
pecially in statistical estimation of reaction rates. This difficulty is a
mplified when direct data on system states are unavailable\, and one only
has access to a random sample of product formation times. To address this\
, we proceed in two stages. First\, under certain technical assumptions ak
in to those made in the Quasi-steady-state approximation (QSSA) literature
\, we prove two asymptotic results: a stochastic averaging principle that
yields a lower-dimensional model\, and a functional central limit theorem
that quantifies the associated fluctuations. Next\, for statistical infere
nce of the parameters of the original MM reaction network\, we develop a m
athematical framework involving an interacting particle system (IPS) and p
rove a propagation of chaos result that allows us to write a product-form
likelihood function. The novelty of the IPS-based inference method is that
it does not require information about the state of the system and works w
ith only a random sample of product formation times. We provide numerical
examples to illustrate the efficacy of the theoretical results. Preprint:
https://arxiv.org/abs/2409.06565\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/60/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Kirstii Badcock (University of Essex)
DTSTART:20241121T120000Z
DTEND:20241121T130000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/61
DESCRIPTION:Title: SMSAS and Impact: Ensuring your research makes a differe
nce\nby Kirstii Badcock (University of Essex) as part of (ED-3S) Essex
Data Science Seminar Series\n\nLecture held in CTC 3.02.\n\nAbstract\nI w
ill give a short overview of national research assessments\, why impact is
important and how it is scored in the REF drawing on examples of high sco
ring impact case studies taken from REF21 for Unit of Assessment 10.\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/61/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Yi Xia (University of Essex)
DTSTART:20250116T120000Z
DTEND:20250116T130000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/62
DESCRIPTION:Title: A Numerical Investigation of Non-Convex Reinsurance Prob
lems Using a Method of Homotopy Optimization with Perturbations and Ensemb
les\nby Yi Xia (University of Essex) as part of (ED-3S) Essex Data Sci
ence Seminar Series\n\nLecture held in CTC 3.02.\nAbstract: TBA\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/62/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Stuart McDonald (Longevity and Demographic Insights)
DTSTART:20250130T120000Z
DTEND:20250130T130000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/63
DESCRIPTION:Title: Mortality in the wake of the pandemic\nby Stuart McD
onald (Longevity and Demographic Insights) as part of (ED-3S) Essex Data S
cience Seminar Series\n\nLecture held in CTC 3.02.\n\nAbstract\nStuart McD
onald will discuss the role of actuaries in proving modelling and insights
and addressing misinformation during the Covid-19 pandemic. He will show
the impact of the pandemic and resulting NHS pressures on UK mortality rat
es\, and the challenge this presents for projecting future mortality.\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/63/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Dr Rishideep Roy (University of Essex)
DTSTART:20250313T120000Z
DTEND:20250313T130000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/64
DESCRIPTION:Title: Real-time within game forecasting in football\nby Dr
Rishideep Roy (University of Essex) as part of (ED-3S) Essex Data Science
Seminar Series\n\nLecture held in CTC 3.02.\n\nAbstract\nWe employ a Baye
sian methodology to predict the results of soccer matches in real-time. Us
ing sequential data of various events throughout the match\, we utilise a
multinomial probit regression in a novel framework to estimate the time-va
rying impact of covariates and to forecast the outcome. English Premier Le
ague data from eight seasons are used to evaluate the efficacy of our meth
od. Different evaluation metrics establish that the proposed model outperf
orms potential competitors inspired by existing statistical or machine lea
rning algorithms. Additionally\, we apply robustness checks to demonstrate
the model’s accuracy across various scenarios.\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/64/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Dr Mengchu Li (University of Birmingham)
DTSTART:20250522T110000Z
DTEND:20250522T120000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/65
DESCRIPTION:Title: Differential privacy analysis of Langevin algorithms
\nby Dr Mengchu Li (University of Birmingham) as part of (ED-3S) Essex Dat
a Science Seminar Series\n\nLecture held in CTC 3.02.\n\nAbstract\nIn this
talk\, I will introduce the concept of differential privacy\, the prevail
ing framework for developing statistical procedures while quantifying the
amount of privacy offered to each individual in the data set. Differential
privacy guarantees are often achieved by injecting noise into determinist
ic algorithms\, and this fact makes a large class of sampling algorithms n
aturally private without any modifications. I will focus on the simplest u
nadjusted Langevin algorithm and discuss several attempts to characterise
its privacy guarantees under the differential privacy framework.\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/65/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Romina Hashami (University of Essex)
DTSTART:20251016T120000Z
DTEND:20251016T130000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/66
DESCRIPTION:Title: Can News Predict the Direction of Oil Price Volatility?
A Language Model Approach with SHAP Explanations\nby Romina Hashami (U
niversity of Essex) as part of (ED-3S) Essex Data Science Seminar Series\n
\nLecture held in 4.311.\n\nAbstract\nFinancial markets can be highly sens
itive to news\, investor sentiment and economic indicators\, leading to im
portant asset price fluctuations. In this study we focus on crude oil\, d
ue to its crucial role in commodity markets and global economy. Specifical
ly we are interested on understanding the directional changes of oil price
volatility\, and for this purpose\, we investigate whether news alone --
without incorporating traditional market data -- can effectively predict t
he direction of oil price movements. Using a decade-long dataset from Eiko
n (2014–2024)\, we develop an ensemble learning framework to extract pre
dictive signals from financial news. Our approach leverages diverse sentim
ent analysis techniques and cutting-edge language models\, including FastT
ext\, FinBERT\, Gemini\, and LLaMA\, to capture market sentiment and textu
al patterns. We benchmark our model against the Heterogeneous Autoregressi
ve (HAR) model and assess statistical significance using the McNemar test.
Notably\, while most sentiment-based indicators do not consistently outpe
rform HAR\, the raw news count emerges as a robust predictor. Among embedd
ing techniques\, FastText proves most effective for forecasting directiona
l movements. Furthermore\, SHAP-based interpretation at the word level rev
eals evolving predictive drivers across market regimes: pre-pandemic empha
sis on supply-demand and economic terms\; early pandemic focus on uncertai
nty and macroeconomic instability\; post-shock attention to long-term reco
very indicators\; and war-period sensitivity to geopolitical and regional
oil market disruptions. These findings highlight the predictive power of n
ews-driven features and the value of explainable NLP in financial forecast
ing.\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/66/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Dr Peter Young (King's College London)
DTSTART:20251106T130000Z
DTEND:20251106T140000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/67
DESCRIPTION:Title: Argumentation Theory and its Applications in Data Scienc
e\nby Dr Peter Young (King's College London) as part of (ED-3S) Essex
Data Science Seminar Series\n\nLecture held in 4.311.\n\nAbstract\nArgumen
tation theory is a branch of artificial intelligence (AI) that models how
claims and arguments interact and how conflicting viewpoints can be resolv
ed. While rooted in formal logic\, argumentation theory has found new rele
vance in data science\, offering tools for modelling and analysing debates
\, explainable AI\, and decision-making with uncertain and contradictory i
nformation.\n\nIn this talk\, I will introduce the foundational concepts o
f argumentation theory\, tracing its development from formal logic to comp
utational models. I will then explore its integration with data-driven app
roaches\, highlighting my recent work in the analysis of social media disc
ourse and in information synthesis for portfolio optimisation. These examp
les illustrate how ideas from argumentation can complement approaches in d
ecision-making and statistical learning.\n\nThis talk is aimed at research
ers interested in the intersection of AI\, data science\, and computationa
l social science\, and does not assume any prior background in logic or ar
gumentation.\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/67/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Yi Xia (University of Essex)
DTSTART:20251120T130000Z
DTEND:20251120T140000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/68
DESCRIPTION:by Yi Xia (University of Essex) as part of (ED-3S) Essex Data
Science Seminar Series\n\nLecture held in 4.311.\nAbstract: TBA\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/68/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Dr Dane Grundy (Aviva)
DTSTART:20251127T130000Z
DTEND:20251127T140000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/69
DESCRIPTION:by Dr Dane Grundy (Aviva) as part of (ED-3S) Essex Data Scienc
e Seminar Series\n\nLecture held in 4.311.\nAbstract: TBA\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/69/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Dr Yang Lu (University of Concordia)
DTSTART:20251204T140000Z
DTEND:20251204T150000Z
DTSTAMP:20260404T111410Z
UID:Essex-DataScience/70
DESCRIPTION:by Dr Yang Lu (University of Concordia) as part of (ED-3S) Ess
ex Data Science Seminar Series\n\nLecture held in 4.311.\nAbstract: TBA\n
LOCATION:https://stable.researchseminars.org/talk/Essex-DataScience/70/
END:VEVENT
END:VCALENDAR