CIMPA School Final Report

Fill in this report with your favorite text editor. The final document has to be converted to a pdf file.

Title of the school: Mathematical Methods in Data Analysis

Dates: July 18-29
Location: Tirana, Albania

I. Summary
General presentation of the event: goals achieved, issues, links you have created in the country or the
region, follow-up of your action.

This was a two weeks summer school introducing topics in Mathematics related to data science. There were
5 main courses (4 hours of lectures + two hours of discussion/exercises for each course), two half courses
(2+1 hour), and 7 special lectures (1 hour). The first week concerned Algebraic Statistics, Convex
Geometry, and Algebra Software in Data Analysis. The second week involved Numerical and Theoretical
approaches to Inverse Problems, Network Data, and X-ray Tomography.

Apart from speakers and organizers, there were about 65 participants most of which were from Albania,
Kosovo, Turkey, North Macedonia, Philippines, Malaysia, Algeria, Egypt, India, and included other
countries such as England, France, Germany, Sweden, and USA. This was the first time for many of the
participants from Albania, Kosovo, and other developing countries to have a first hand experience with math
and professors outside their university curriculum. The participants proved that students in these countries
are eager to embrace such opportunities.

Students learned about career opportunities. There were two panel sessions dedicated to sharing advice and
future career opportunities for the participants. In particular, the director of CIMPA Christophe Ritzenthaler
shared many concrete opportunities and links with the students. Each day, we had two 30 minutes breaks for
students and professors to interact. The lunch was provided by local funding for all the participants in the
school. This was another opportunity for students to make connections with each-other and professors. We
also had an excursion, and two gala dinners to help connect the participants with each other in friendly
environments.

The school organization proved to be a successful first big collaboration in math between the diaspora of
Albania and professors in Tirana. Many professors, not only the ones involved in this school, expressed
interest in following up with other events, possibly of a smaller size. The hosting organization, the Faculty
of Natural Sciences at the University of Tirana, supported this event unconditionally, and is open to future
ones. A special support and interest for future events came from the Albanian Academy of Sciences.

The organizers successfully learned where and how to find sponsors for such events in Albania and feel
comfortable organizing such events in the future. In particular, Florion Cela, an assistant professor at the
University of Tirana and local organizer, was the main person able to attract sponsors such as the City Hall
of Tirana, Tirana European Youth Capital ‘22, and the institutions from Kosovo NinetyOne and The New
York School of Sciences. All these institutions expressed satisfaction with the results of the school and are
open to contributing and collaborating for future events. A regional conference/workshop is being discussed.
II. Scientific content
List here the activities and lectures which took place with the names of teachers.

MAIN COURSES
Algebra Software in Data Analysis
Paul Breiding (Max Planck Institute for Mathematics in the Sciences)
The developments in science and technology grow in parallel. Mathematical models, especially those used in data analysis and
numerical analysis are in need of software to implement, test, and validate results. In this course we aim to complement the theoretical
and practical developments with software learning as a tool to practice and produce results. The goal is to start with basic
introductory concepts in Julia. The course will be followed by training sessions where the participants will have the opportunity to
practically try the tools and packages, use them on a variety of tasks, ask questions, and develop a good understanding of the software.

Maximum Likelihood Estimation for Toric Varieties

Eliana Duarte (Centro de Matemática Universidade do Porto)
The goal of these lectures is to illustrate how the maximum likelihood estimation problem for log-linear models can be studied from
an algebraic geometric perspective. In the first lecture we will start by establishing that discrete exponential families are in fact toric
varieties. We will also focus on understanding several graphical models such as undirected graphical models, discrete Bayesian
networks and staged tree models. In the second lecture we will address the maximum likelihood estimation problem for this model
class, define the maximum likelihood degree (MLdegree) of a model and take a look into the classification of discrete statistical models
with MLdegree equal to one. In the last lecture we will go over the classification of log-linear models with MLdegree one in
dimension two and present some open problems and conjectures.

New Approaches to Inverse Scattering

Houssem Haddar (INRIA, Center of Saclay Ile de France)
In the past thirty years the field of inverse scattering theory has become a major theme of applied mathematics with applications to
such diverse areas as medical imaging, geophysical exploration and nondestructive testing. The growth of this field has been
characterized by the realization that the inverse scattering problem is both nonlinear and ill-posed, thus presenting particular
problems in the development of efficient inversion algorithms. Although linearized models continue to play an important role in
many applications, the increased need to focus on problems in which multiple scattering effects can no longer be ignored has led to
the nonlinearity of the inverse scattering problem playing a central role. In addition, the possibility of collecting large amounts of data
over limited regions of space has led to the situation where the ill-posed nature of the inverse scattering problem becomes a problem
of central importance. Initial efforts to deal with the nonlinear and ill-posed nature of the inverse scattering problem focused on the
use of nonlinear optimization methods. Although efficient in many situations, their use suffers from the need for strong a priori
information in order to implement such an approach. In addition, recent advances in material science and nanostructure fabrications
have introduced new exotic materials for which full reconstruction of the constitutive parameters from scattering data is challenging
or even impossible. In order to circumvent these difficulties, a recent trend in inverse scattering theory has focused on the
development of a new approach, in which the amount of a priori information needed is drastically reduced but at the expense of
obtaining only limited information of the scatterers.

We present an overview of the aforementioned new ideas in the context of solving the inverse scattering problem for anisotropic
inhomogeneous media. Our approach exploits properties of the linear scattering operator to decode non-linear information about the
scattering medium, yielding mathematically justified and computationally simple reconstruction algorithms. We first show that the
support of the scattering media can be rigorously characterized from the range of the scattering operator. The mathematical study of
this operator is based on the analysis of a non-selfadjoint eigenvalue problem known as the transmission eigenvalue problem.
Transmission eigenvalues relate to interrogating frequencies for which there is an incident field that does not scatter. We show that
the transmission eigenvalues can be determined from the scattering data and provide information on the constitutive properties of
the scattering media. Numerical examples will be presented to confirm the viability of our reconstruction algorithms. The course is
intended to be self contained. The first couple of lectures will be an introduction to set up the problem formulation and the
background information needed. The following lectures will develop research questions and introduce more advanced research in the
field.

0/1-Polytopes in Algebra, Discrete Geometry and Applications

Tim Römer (Universität Osnabrück) and Yulia Alexander (University of California-Berkeley)
There exist beautiful challenges related to (high dimensional) 0/1-polytopes in algebra, discrete geometry, and optimization as well as
applications of results achieved in areas like data analysis. Instances of such situations arise for example in the study of questions
related to Karp's list of 21 NP-complete problems from combinatorial optimization. An explicit example for this is MaxCut and
associated cut objects like cut polytopes and -algebras. In these lectures we discuss recent results and open questions of 0/1-polytopes
from these different perspectives. A goal is to understand the interplay between geometric properties of 0/1-polytopes, algebraic
properties of associated toric algebras and interpretations of results obtained in more applied areas.

Invitation to X-ray Tomography

Samuli Siltanen (University of Helsinki)
X-ray tomography is a classical imaging method, based on analyzing the weakening of X-rays sent through a physical body along
various directions. It is used daily for medical imaging in hospitals, and in industrial non-destructive testing applications.
Mathematically, the tomographic imaging task is the ill-posed inverse problem of recovering a non-negative function from a
collection of its line integrals. Things get really interesting when the imaging directions are restricted, or the target is moving during
imaging, or when the radiation is emanating from inside the target instead of external probing. This course builds up the needed
mathematical machinery starting from basic linear algebra and ending up with currently active research topics in tomographic
imaging. The methods are applied to actual measured data created for the course and available freely online.

SHORT COURSES
Efficient Krylov subspace methods for uncertainty quantification
Julianne Chung (Virginia Tech)
Uncertainty quantification for linear inverse problems remains a challenging task, especially for problems with a very large number of
unknown parameters (e.g., dynamic inverse problems), for problems where computation of the square root and inverse of the prior
covariance matrix are not feasible, and for hierarchical problems where the mean is not known a priori. This work exploits Krylov
subspace methods to develop and analyze new techniques for large-scale uncertainty quantification in inverse problems. We assume
that generalized Golub-Kahan based methods have been used to compute an estimate of the solution, and we describe efficient
methods to explore the posterior distribution. We present two methods that use the preconditioned Lanczos algorithm to efficiently
generate samples from the posterior distribution. Numerical examples from dynamic photoacoustic tomography and atmospheric
inverse modeling, including a case study from NASA's Orbiting Carbon Observatory 2 (OCO-2) satellite, demonstrate the
effectiveness of the described approaches.

Learning Methods for Network Data

Evin Isufi (Delft University of Technology)
Network data are everywhere with prominent examples those generated by social, economic, biological, or sensor networks. Network
data are complex and irregular, therefore, they do not lend themselves to standard processing tools. Recent years have seen a surge of
interest to develop machine and deep learning methods for network data. In this short training course, we will cover the fundamental
concepts behind these methods, focusing on graph signal processing and graph neural networks (GNN). We will also touch upon the
interplay between the two and discuss recent results on the stability of GNNs to perturbations. Finally, we will provide a handful of
blueprint tasks of how to categorize several network-based learning tasks arising in different domains.

Computational Uncertainty Quantification for Inverse Problems in Python

Jakob Sauer Jørgensen (Technical University of Denmark)
In this short training course, we will present a new Python software package for modeling and computations of uncertainty
quantification (UQ) for inverse problems such as image deblurring and computed tomography. The goal of the software is to provide
an easy-to-use platform that allows non-experts in UQ and Bayesian inversion to perform UQ analysis of inverse problems. At the
same time, the software allows expert users full control of the models and methods. After a brief introduction to UQ, we demonstrate
the main features of the software via examples and hands-on exercises using the built-in test problems. Then we describe more
advanced features, and we demonstrate how to work with one’s own inverse problem in the software package.

SPECIAL LECTURES

Building machine learning models that generalize: lessons from the literature on what works
and what doesn't
Visar Berisha (Arizona State University)
A fundamental question that algorithm designers must answer as they develop a machine learning model is “How well will this model
perform when it's deployed in the real world?” The standard approach to answering this question in the current ML literature is
cross-validation. We provide evidence across a number of application areas that this approach likely leads to overoptimistic estimates
of the true generalization performance of machine learning models. Using examples from digital health and geometric intuition, we
explain why it’s difficult to estimate how well a machine learning model will perform when deployed. Then we suggest several
mitigation strategies for how to build robust models that generalize to real-world conditions.

Transmission Eigenvalues and Inverse Problems in Euclidean and Hyperbolic Geometry

Fioralba Cakoni (Rutgers University)
The transmission eigenvalue problem is at the heart of inverse scattering theory for inhomogeneous media. Transmission eigenvalues
are related to interrogating frequencies for which there is an incident field that doesn’t scatter by a given medium. They are
eigenvalues of a non-selfadjoint eigenvalue problem with a deceptively simple formulation, namely two elliptic PDEs in a bounded
domain that share the same Cauchy data on the boundary, but presenting a perplexing mathematical structure. The connection
between transmission eigenvalues and non-scattering energies is well studied in the Euclidean geometry, where in special cases these
eigenvalues appear as zeros of the scattering matrix. Transmission eigenvalues can be determined from scattering data and can be used
in imaging inhomogeneous media.
In this presentation, we first discuss how transmission eigenvalues appear in scattering by various structures in Euclidean setting, and
review some of the state-of-the-art results and interesting open problems. Then, we extend the concept of transmission eigenvalues
to the scattering theory for automorphic forms on fundamental domains generated by discrete groups acting on the hyperbolic upper
half complex plane. For arithmetic groups, using existing deep results from analytic number theory, we reveal that the zeros of the
scattering matrix, consequently non-scattering energies, are directly expressed in terms of the zeros of the Riemann zeta function. We
provide Weyl’s asymptotic laws for the transmission eigenvalues in this setting along with estimates on their location in the complex
plane. Finally, we present some future prospects in this direction.

Quasi-Independence Models with Rational Maximum Likelihood Estimator

Jane Ivy Coons (University of Oxford)
Let X and Y be random variables. Quasi-independence models are log-linear models that describe a situation in which some states of
X and Y cannot occur together, but X and Y are otherwise independent. We characterize which quasi-independence models have
rational maximum likelihood estimator, or MLE, based on combinatorial features of the bipartite graph associated to the model. In
this case, we give an explicit formula for the maximum likelihood estimate. We also show that if a log-linear model has rational MLE,
then so do all of its facial submodels

Nonlinear algebra and Nonlinear Statistics for Networks

Sonja Petrović (Illinois Institute of Technology)
The ubiquity of network data in the world around us does not imply that the statistical modeling and fitting techniques have been
able to catch up with the demand. This series of lectures will discuss some of the basic modeling questions that every statistician
knows are fundamental, some of the recent advances toward answering them, and the challenges that remain. The specific focus of
the lectures will be on goodness of fit testing for random graph models.

The lectures will be focused around the testing framework for graphs that is based on combinatorics of hypergraphs. More broadly,
we will summarize a few lines of research that are intimately connected to discrete mathematics and computer science, where
sampling algorithms, hypergraph degree sequences, and polytopes play a crucial role in the general family of statistical models for
networks called exponential random graph models

Cloaking via change of variables in Electrical Impedance Tomography

Michael Vogelius (Rutgers University)
This talk is a survey of the fascinating topic of invisibility cloak in imaging. The goal of cloaking is to render an object invisible to
detection from electromagnetic energy by surrounding the object with a specially engineered “metamaterial” that redirects
electromagnetic waves around the object. We show how to cloak an object against detection from impedance tomography, an imaging
technique of much recent interest, though the mathematical ideas apply to much more general forms of imaging via waves.

Spectral Graph Analysis Revealed from Matrix Functions

Eglantina Xhaja (University of Tirana)
Networks are all around us and their properties are studied through their associated matrices: the adjacency matrix, the graph
Laplacian. Both matrices have been well studied from an algebraic point of view, but when it comes to real networks, the matrix
dimension increases extremely, which leads to thinking about faster processing. A useful notation in the analysis of complex networks
based on spectral analysis is the communicability between nodes, temporal communicability and in general the overall
communicability of the network, which are expressed as the exponential of the adjacency matrix and its invariants in the cases of
undirected graphs and as hyperbolic cosines and sinus of the adjacency matrix. There exist a lot of methods for the calculation of the
exponential of a matrix, and one of the best is scaling and squaring [Higham, 2005]. In this presentation we will use a computational
strategy proposed by Benzi & Simoncini [2017], in the evaluation of the exponential of the matrix, which is based on the
Kronecker product, which is more appropriate for large dimensions. The matrix functions have been well studied from an algebraic
point of view, but when it comes to real network, the matrix dimensions increase extremely, which leads to think about efficient
method for computing the exponential of large and rare matrices, and the methods of Krylov subspaces are a powerful mechanism in
this contents, a matrix-free method that uses only matrix-vector products. The Laczos method combined with Grand-Schmidt
orthogonalization to compute them, specially dealing with case when the problem is will- conditioned.The numerical tests for real
networks (social networks, communication networks and genetic networks) with matrix dimensions, which vary from thousands to
millions, are performed in Matlab R2011b and are presented cases were shared memory environment could be used instead.

Men/Women Speakers: 8 of the speakers were men and 7 were women.

III. Participants
List here all participants, including local participants, foreign participants without CIMPA support and
foreign participants supported by CIMPA. Please mention possible strong participants (with their email)
that you think CIMPA should follow (CIMPA wishes to serve as a recommendation center for students
applying to doctoral or postdoctoral grants. Also these persons may be natural contacts for CIMPA in the
future).

The following contains the names of all participants. The bolded names at the end of the list are
recommendations, together with their current positions.
 Role/Positio
n (if blank is
Name Gender Affiliation Email a participant)

troemer@uni-osnabru
Tim Romer M University of Osnabrück eck.de Speaker

Sonja Petrovic F Illinois Institute of Technology [email protected] Speaker

Elvin Isufi M Delft University of Technology [email protected] Speaker

Eliana Duarte F University of Porto [email protected] Speaker

Samuli Siltanen M University of Helsinki [email protected] Speaker

Houssem.Haddar@inri
Houssem Haddar M INRIA a.fr Speaker

[email protected].
Jane Ivy Coons F University of Oxford uk Speaker

Visar Berisha M Arizona State University [email protected] Speaker

Julianne Chung F University of Virginia [email protected] Speaker

Jakob Jorgensen M Technical University of Denmark [email protected] Speaker

[email protected]
Michael Vogelius M Rutgers University s.edu Speaker

Max Planck Institute MiS & [email protected]

Paul Breiding M University of Osnabrück g.de Speaker

[email protected] Speaker/org
Fioralba Cakoni F Rutgers University du anizer

[email protected]
Yulia Alexander F University of California-Berkeley du Speaker

Christophe
Ritzenthaler M CIMPA & Rennes 1 University [email protected] CIMPA repr

eglantina.kalluci@fshn Speaker/org
Eglantina Xhaja F University of Tirana .edu.al anizer

Aida Maraj F University of Michigan [email protected] Organizer

Mirjeta Pasha F Arizona State University [email protected] Organizer

besiana.hamzallari@fs
Besiana Cobani F University of Tirana hn.edu.al Organizer

ferdinand.bego@fshn.
Ferdinand Bego M University of Tirana edu.al Organizer

Florion Cela M University of Tirana Organizer

[email protected].
 al

[email protected]
Elisabeta Peti F University of Tirana u.al

University of the Philippines Los [email protected].

Egarguin Neil Jerome M Banos ph

Middle East Technical

Mustafa Kuturuk M University [email protected]

Esra Emet F Karadeniz Technical University [email protected]

Javid Ali M Aligarh Muslim University [email protected]

Universiti Teknologi mudasar_20000296@ut

Mudasar Zafar M PETRONAS p.edu.my

KTH Royal Institute of

Pratik Misra M Technology [email protected]

Ben Hollering M Max Planck Institute MiS [email protected]

josephcummings03@g
Joseph Cummings M University of Notre Dame mail.com

ilir.berisha19@student.
Ilir Berisha M University of Prishtina uni-pr.edu

Huseyin Yildirim M The University of Essex [email protected]

University of Applied Sciences [email protected]

Valdete Loku F Ferizaj m

Naim Braha M University of Prishtina [email protected]

Max Planck Institute MiS & [email protected]

Elima Shehu F University of Osnabrück .de

KTH Royal Institute of danai.deligeorgaki18@

Danai Deligeorgaki F Technology alumni.imperial.ac.uk

[email protected]
Burcu Aydogan F RWTH Aachen University m

Mandala von mvonwestenho@uni-os

Westenholz F University of Osnabrück nabrueck.de

Conor Stadenford M Arizona State University [email protected]

Fjoralba Xhaferi F Polytechnic University of Tirana [email protected]

[email protected]
Arnisa Sokoli F University of Tirana u.al

Anxhela Gjecka F University of Tirana [email protected]

Sauda Nerjaku F University of Tirana

sauda.nerjaku@gmail.
 com

anxhelo.shehu99@gma
Anxhelo Shehu M Polytechnic University of Tirana il.com

Mediterranean University of [email protected]

Teuta Xhindi F Albania m

American International brikena.vrioni@yahoo.

Brikena Vrioni F University com

aleksander.kovaci@uet.
Sander Kovaci M Polytechnic University of Tirana edu.al

raimondadervishi@yah
Rajmonda Dervishi F Polytechnic University of Tirana oo.com

Hedera Pema F Institution Harry. T. Fultz [email protected]

[email protected]
Ornela Darova F University of Pennsylvania du

[email protected].
Arjan Tushaj M University of Tirana al

arben.baushi@univlora.
Arben Baushi M University of Vlora edu.al

Orgest Zaka M Agricultural University [email protected]

Besjana Jacaj F Epoka University [email protected]

[email protected]
Klaudio Peqini M University of Tirana u.al

American International
Ervin Hoxha M University

Hysen Ndregjoni M Epoka University

Arban Uka M Epoka University

jon.dulja@fshnstudent.
Jon Dulja M University of Tirana info Bachelor

anwalker@willamette. Masters
Alexandra Walker F Willamette University edu student

Université Paris 1 [email protected] Masters '

Doris Allamani F Panthéon-Sorbonne om Student

sevdebusra.bayrak@gm Masters '

Sevde Büşra Bayrak F Boğaziçi University ail.com Student

[email protected] Masters '

Erion Hamiti M University of Prishtina ni-pr.edu Student

Florent Zymeri M University of Prishtina

florent.zymeri@kosova. Masters '
 maarifschools.org Student

[email protected] Masters '

Ilir Destani M University of Prishtina om Student

Masters '
Fiton Hoxha M University of Prishtina [email protected] Student

[email protected] Masters '

Abdullah Sylejmani M University of Prishtina om Student

Ss. Cyril and Methodius alban.ismaili98@gmail. Masters '

Alban Ismaili M University com Student

danielabardhi35@gmail Masters '

Daniela Bardhi F University of Tirana .com Student

klaudia.ndreca2@fshns Masters '

Klaudia Ndreca F University of Tirana tudent.info Student

[email protected] Masters '

Patris Sade F University of Tirana m Student

megigerboti17@gmail. Masters '

Megi Gërboti F University of Tirana com Student

vera.hoxha@fshnstude Masters '

Vera Hoxha F University of Tirana nt.info Student

memjondecka@gmail. Masters '

Memjon Deçka M University of Tirana com Student

adrianxhindoli@gmail. Masters '

Adrian Xhindoli M University of Tirana com Student

hedga.hida@fshnstude Masters '

Hedga Hida F University of Tirana nt.info Student

adilegundogdu34@gma Masters '

Adile Gündoğdu F Boğaziçi University il.com Student

University of Tirana mucobegajonida@gm Masters '

Jonida Muçobega F ail.com Student

Master of
Science
Finished,
Teaching
Eriselda Goga F Epoka University [email protected] Assistant

Master of
Science
Finished,
agresa.qosja1@gmail. Teaching
Agresa Qosja F University of Tirana com Assistant

Master of
Science
National Agency of Employment Finished,
Ina Hila F and Skills [email protected]
Part time
 lecturer and
Industry

rahmajerbi1995@gmai
Rahma Jerbi F University of Sfax l.com PhD student

Ph.D.
Student and
[email protected] Lecturer at
Xhilda Dhamo F University of Tirana u.al UT

Ph.D.
Student and
[email protected] Teach at
Fizeleni Lekli F Polytechnic University of Tirana m UPT

mohamedn6666@gmai
Ali Mohamed Naser M Helwan University l.com PhD student

[email protected]. Ph.D.
Gazi Ozdemir M Izmir Institute of Technology tr Student

[email protected] Ph.D
Destiny Lutero F Ateneo de Manila University eo.edu student

Universiti Teknologi abidahussain490@yaho

Abida Hussain F PETRONAS o.com PhD student

Polytechnic University of Tirana [email protected] Ph.D.

u.al Student and
Teach at
Epoka and
Redjola Manaj F UT

Ermal Feleqi M University Ismail Qemali [email protected] Assoc. Prof.

Men and Women Participants: 43 of participants were men and 47 were women.

IV. Financial Report

Provide here the list of all grants received for the organization of your school. Then give details about
expenses made with CIMPA funding.

The following are the grants received. Note that each of them was in charge of parts of the school, described
briefly below.

1. Faculty of Natural Sciences, University of Tirana –10.000 Euro, used for lodging + breakfast of
participants and drinks and snacks during breaks (twice a day).
2. Institut national de recherche en sciences et technologies du numérique INRIA –2.500 Euro,
used for lodging + breakfast of participants
3. NinetyOne and The New York School of Sciences –5.000 Euro, used for two gala dinners for all
the participants
4. The Albanian Academy of Science –5.000 Euro, provided lodging and breakfast for the speakers
and two dinners with the speakers and organizers.
5. Tirana European Youth Capital ‘22 –4.000 Euro, provided lunch catering in first week for all the
participants
 6. City Hall of Tirana – 4.000 Euro, provided lunch catering in second week for all the participants
7. CIMPA – 12.000 Euro, were used for
● Lodging of 10 CIMPA participants
● Dinners for 10 participants
● Flight tickets (mostly full unless the CIMPA participants had other funds available and the
flight was expensive) of 10 CIMPA participants and 1 non-CIMPA participant whose lodging
was offered by the locals.
● Bus ticket for 5 students from Kosovo and 1 from North Macedonia
● Entrance tickets for “The House for Leaves”, “BunkArt 1”, and the local transportation by
bus and Teleferik for all the participants that attended the excursion on the weekend.
● Printing expenses such as nametags, posters, handbag, notebook, pen…
In addition, all speakers and non-local organizers paid with their own fundings.

V. Pictures
Please include pictures of school and social activities.