Toggle description

Prof Emilio Carrizosa has been invited to offer a PhD course on Optimization and Machine Learning within the PhD program in Information engineering , Univ of Florence. The course will be held on line from Monday June 7th to Thursday June 10, from 16.00 to 18.00 CEST. Interested PhD students can send me an email in order to get access to the lectures.

Mathematical Optimization is at the core of many Machine Learning problems in classification, regression and dimensionality reduction, amon others. An important challenge is to make classification and prediction algorithms more interpretable, in the sense that we should know which attributes, and at which extent, contribute in the prediction.
Mathematical Optimization allows us to pose in a natural way the multiobjective problem of optimizing the performance and, at the same time, the number of attributes or measurement costs.

In this course we will illustrate the use of Mathematical Optimization strategies in different problems, such as dimensionality reduction (sparse PCA), sparse linear models with performance constraints, cost-sensitive Support Vector Machines with performance constraints or functional data, sparse classification and regression (ensembles of) trees, interpretable clustering, etc., with special focus on the methods developed by the research group in Optimization in IMUS, the Institute of Mathematics of the University of Seville

Toggle description

Since 2007, NATCOR has been providing courses for PhD students and early career researchers working in Operational Research and related areas, such as Computer Science, Industrial Engineering and Business Analytics. The NATCOR team now includes members of fourteen UK universities, and it is endorsed by the Engineering and Physical Sciences Research Council (EPSRC), the Operational Research Society (ORS) and the European Association of Operational Research Societies (EURO).

The following course will be take place, via MS Teams, on 12th - 16th April

Stochastic Modelling Course

Stochastic Modelling is concerned with using probability concepts and techniques for capturing uncertainty in order to describe situations, predict performance and support decision making. The Operational Research literature abounds with applications of Stochastic Modelling, with Healthcare, Transportation, Computing and Communications, Business and Finance being just a few examples. In the recent years, Stochastic Modelling has become a key component of interdisciplinary research between Operational Research and Statistics and Machine Learning. This course will present some of the theory behind such modelling processes, but consideration will also be given to applications by means of case studies. The topics covered are: Stochastic Processes, Queueing Systems and Networks, Maintenance and Reliability, Inventory Control and Revenue Management. This course will run fully online, in a blended way including interactive sessions, and there will also be plenty of time for informa!
l interactions and networking.

Course Leader:
Dr Peter Jacko (Lancaster University)

Lecturers:
Dr Rob Shone (Lancaster University)
Dr Chris Kirkbride (Lancaster University)
Dr David Worthington (Lancaster University)
Prof Adam Letchford (Lancaster University)
Dr Dong Li (Loughborough University)
Prof Shaomin Wu (University of Kent)

Lectures will be complemented by invited case study presentations by Richard Fussey (Blue Yonder) and Dr. Bin Liu (University of Strathclyde)

Registration will close at 12 noon on 26th March 2021 so register [http://www.natcor.ac.uk/register/] now to reserve your place!

Detailed descriptions of our other courses, and details about registration, assessment and accreditation may be found on the website ( http://www.natcor.ac.uk [http://www.natcor.ac.uk/]).

Toggle description

Cari tutti,



nella settimana dal 22 al 26 marzo prossimi, nell’ambito del Dottorato di Ricerca in Matematica e Informatica dell’Università della Calabria (Dipartimento di Matematica e Informatica), terrò un corso dal titolo “Optimization Models for Machine Learning”, 4 CFU, 12 ore.



Obiettivo del corso sarà quello di presentare alcuni modelli di ottimizzazione finalizzati al Machine Learning, con riferimento alla classificazione supervisionata, non supervisionata e semi-supervisionata. Il corso inoltre è “self-contained”: quindi la prima parte sarà introduttiva e dedicata alla presentazione dei principali concetti dell’Ottimizzazione Matematica. Una piccolo spazio sarà anche dedicato ai problemi di Multiple Instance Learning. Ecco il programma nel dettaglio:



PART I: Introduction to Numerical Optimization

The optimization problems;
The min-max problems;
Convexity;
Global and local minima;
Optimality conditions;
The Wolfe dual problem.
PART II: Numerical Optimization and Machine Learning

Introduction to Machine Learning;
Optimization and pattern classification;
Optimization models for supervised classification;
Linear separation;
Spherical separation;
Polyhedral separation;
Support Vector Machine;
The kernel trick;
Proximal Support Vector Machine;
Spherical separation with margin;
Optimization models for unsupervised classification;
The non-smooth clustering optimization model;
Optimization models for semi-supervised classification;
Transductive Support Vector Machine;
Semi-supervised spherical separation;
Semi-supervised polyhedral separation;
Multiple Instance Learning;
Instance-space, bag-space and embedding-space approaches;
Support Vector Machine for Multiple Instance Learning;
Evaluation of a classifier;
Accuracy;
Sensitivity;
Specificity;
Precision;
F-score;
Model selection: cross validation and leave-one-out strategies.


Le lezioni si svolgeranno sulla piattaforma Teams e sono rivolte a dottorandi e/o studenti magistrali. Se qualcuno è interessato a seguire le lezioni, può contattarmi all’indirizzo antonio.fuduli@unical.it.



Cordiali saluti a tutti,

Antonio

Toggle description

Cari Colleghi,

dal 15 al 19 marzo, nell'ambito del Dottorato DIBRIS in Informatica e
Ingegneria dei Sistemi, si svolgerà il Corso

"Strategic Choices: Games and Team Optimization",

tenuto da Lucia Pusillo e Marcello Sanguineti.

Informazioni su

http://phd.dibris.unige.it/csse/index.php/activities-and-credits


http://phd.dibris.unige.it/csse/index.php/activities-and-credits/phd-courses

Cordiali saluti

Marcello Sanguineti

Toggle description

Nelle prossime settimane terro' in videoconferenza un corso di 20 ore dedicato a una panoramica degli algoritmi euristici per problemi di Ottimizzazione Combinatoria.
Il corso si tiene nell'ambito del dottorato di Informatica dell'Universita' degli Studi di Milano.

Le date previste (soggette a possibili cambiamenti) e gli argomenti sono:

15/01/2021 14:00-17:00 Generalita'
18/01/2021 14:00-17:00 Valutazione teorica ed empirica
19/01/2021 14:00-17:00 Euristiche e metaeuristiche costruttive (GRASP, Ant System)
22/01/2021 14:00-17:00 Euristiche di scambio
25/01/2021 14:00-17:00 Metaeuristiche di scambio (ILS,VNS,VND)
27/01/2021 14:00-17:00 Metaeuristiche di scambio (TS, SA)
29/01/2021 14:00-17:00 Metaeuristiche di ricombinazione (SS, PR, algoritmi genetici)

Dettagli e materiali saranno via via pubblicati sulla mia pagina web.

Roberto Cordone

Toggle description

November 26, 2020, h 17:00

Speaker: Francesca Parise

Discussants: Giacomo Como, Daniel Cooney, Mathieu Lauriere.

Click here to access the Virtual Room: http://mailsender.luiss.it/lists/lt.php?id=cRpRCgYHSAUHAAFOVgFdBQs

The papers can be found here:

https://arxiv.org/abs/1802.00080

https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3674691

Information about future seminars can be found here:

https://economiaefinanza.luiss.it/seminari/prossimi-seminari

Toggle description

Timetable: 8 hrs.
The course will be held online (link Zoom will be comunicated).

Calendar of the lectures

Tuesday October 27, 2020, 16:00
Wednesday October 28, 2020, 10:00-12:00
Thursday October 29, 2020, 10:00-12:00 and 15:00-17:00

Speaker: Prof. I. M. Bomze
Title: Conic, especially copositive optimization

Course contents.

Quite many combinatorial and some important non-convex continuous optimization
problems admit a conic representation, where the complexity of solving non-
convex programs is shifted towards the complexity of sheer feasibility (i.e.,
membership of the cone which is assumed to be a proper convex one), while
structural constraints and the objective are all linear. The resulting problem
is therefore a convex one, and still equivalent to some NP-hard problems with
inefficient local solutions despite the fact that in the conic formulation,
all local solutions are global.

Using characterizations of copositivity, one arrives at various
approximations. However, not all of these are tractable with current
technology. In this course, we will address some approaches on which tractable
SDP- or LP-approximations, and also branch-and-bound schemes, may be based.

This way, good tractable bounds can be achieved which serve as quality control
for any primal-feasible algorithm. But which one should be employed?
Complementing above (dual) approach, we will, mainly as one example, address a
classical yet not widely known first-order approach for poly/posynomial
optimization under simplex constraints, embedded in some general optimization
principles for iterative primal methods.