June 2015 - NumPI - Mailing list — Dipartimento di Matematica, UniPI

Optimization & Numerical Analysis Seminars: Flores-Bazan and Locatelli
by Federico Poloni 29 Jun '15

29 Jun '15

Two seminars in optimization, one after the other, are taking place next Tuesday. Speaker1: Fabian Flores-Bazan (University of Concepcion) Title1: Strong duality in nonconvex optimization with a single equality and geometric constraints: applications to QP Time1: Tuesday, June 30; 15:00 Place1: Sala seminari est, informatica Speaker2: Marco Locatelli (Università di Parma) Title2: Some results about convex envelopes Time2: Tuesday, June 30; 15:45 Place2: Sala seminari est, informatica Abstract1: The validity of strong duality for a non convex optimization problem under a single equality and geometric constraints is characterized in a topological and geometric manner. In particular, a hidden convexity of the conic hull of joint-range of the pair of functions associated to the original problem, is obtained. Applications to derive KKT conditions without standard constraints qualification are also discussed. Several examples showing our results provide much more information than those appearing elsewhere, are given. Furthermore, first and second order optimality condition, for a non convex quadratic optimization problem under two quadratic equality constraints, are also presented. Finally, the standard quadratic problem involving a non necessarily polyhedral cone is analyzed in detail. The talk is based on joints works with G. Carcamo and F. Opazo. Abstract2: In this talk we discuss convex envelopes for some bivariate functions over general polytopes and for some $n$-dimensional quadratic functions over the unit simplex. We first give two formulations of convex envelope, a primal and a dual one. Next, we discuss how to derive the convex envelope for some bivariate functions over polytopes, provided that some given assumptions are fulfilled. It will turn out that the formula of the convex envelope can be derived through the solution of a parametric convex program or, equivalently, through the solution of the corresponding KKT system. For what concerns the $n$-dimensional quadratic functions over the unit simplex, we will first associate a graph to the quadratic function, and next we will show that the derivation of the convex envelope is related to the identification of the spanning forests over this graph. === Everyone is welcome! -- --federico poloni Dipartimento di Informatica, Università di Pisa http://www.di.unipi.it/~fpoloni/ tel. +39 050 2213143

2 1

Fwd: Call per data analyst
by Federico Poloni 18 Jun '15

18 Jun '15

Ricevo e inoltro questa offerta di lavoro. -federico -------- Forwarded Message -------- in Zanichelli cercano un data analyst junior a tempo determinato, e una parte della selezione (quella "tecnica") la faccio io. In allegato c'è la call: potreste girarla, per favore, a chiunque possa essere interessato? Gli interessati possono inviare cv e lettera di presentazione al nevio.dubbini(a)gmail.com. (NB il mio indirizzo email non c'è nella call allegata, quindi andrebbe comunicato nel testo dell'email). Grazie, ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Nevio Dubbini mail: nevio.dubbini(a)gmail.com web: sites.google.com/site/neviodubbini ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

1 0

Fwd: [Dipartimento.di] avviso ciclo di seminari - 8, 10, 11 giugno
by Federico Poloni 04 Jun '15

04 Jun '15

In the framework of the research project "Mathematical models and computational methods for complex networks" by the University of Pisa, Paola Boito (Faculté des Sciences et Techniques Université de Limoges) will give three seminars on "Complex networks, functions of matrices and quadrature rules". The abstract is provided below. The seminars will take place in June at the Department of Computer Science of the University of Pisa: -- June 8th, 15:00-16:00, sala seminari est (room 351); -- June 10th, 10:00-11:00, sala seminari est (room 351); -- June 11th, 15:00-16:00, sala seminari est (room 351). Abstract. One cornerstone of our modern society is the use of different kinds of networks. Our cities are connected by a network of streets and railways, telecommunication networks including their wireless, mobile components, the internet, and the World Wide Web build the most important infrastructure for communication and information worldwide. Designing and managing such networks pose challenging algorithmic and numerical problems. These three talks are concerned with the development, analysis and application of numerical linear algebra tools for the quantitative study of important properties of large-scale complex networks, such as centrality, betweenness and communicability measures. Specifically, we design efficient algorithms for the fast approximation of the entries of matrix functions such as the exponential and the resolvent of adjacency matrices and graph Laplacians, as well as their traces. While these matrices are quite sparse, in the case of complex networks they behave very differently from the matrices associated with regular lattices (grids), such as those arising from discretizations of partial differential equations. Hence, there is a need to develop new methods, both computational and analytical, to deal with these very large-scale problems. Potential applications include the analysis of social networks, the study of biological and neurological networks, applications in physics and operations research, and so forth.

1 1

Optimization & Numerical Analysis Seminar: N. Vannieuwenhoven
by Federico Poloni 03 Jun '15

03 Jun '15

Speaker: Nick Vannieuwenhoven (KU Leuven) Title: Towards a perturbation theory for the tensor rank decomposition Date: Thursday, 4 June 2015. Time: 11:00 Place: Sala Seminari Ovest, dipartimento di informatica Abstract: The tensor rank decomposition is a decomposition of a tensor into a linear combination of rank-1 tensors. One of the key advantages of higher-order tensors is that this decomposition is generally unique, which allows for an interpretation of the individual rank-1 terms. Because of this property, the tensor rank decomposition has found application in several domains. For instance, they can be used as a clustering algorithm in an unsupervised learning setting assuming a certain statistical model wherein each of the rank-1 tensors in the decomposition will correspond with a cluster. In the applications where tensor rank decompositions arise, the tensor is usually known only up to some small perturbation error. This raises the following interesting problem: Suppose that the tensor is perturbed by a small quantity, by how much can the individual rank-1 terms change relative to the magnitude of the perturbation? Can it be unbounded? In this presentation, I will explore the definition of a possible condition number that answers these questions up to first order. == Everyone is welcome! -- --federico poloni Dipartimento di Informatica, Università di Pisa http://www.di.unipi.it/~fpoloni/ tel:+39-050-2213143

1 1