Signal Processing and Optimization over Networks

Paolo Di Lorenzo - Assistant Professor in the Department of Information Engineering, Electronics, and Telecommunications at Sapienza University.
Communication systems

Date: April 4th 2018
Location: Eurecom - Eurecom

SUMMARY: Nowadays, large-scale networks are ubiquitous (e.g., Heterogeneous 5G Nets, Internet, Sensor Nets, Social nets, Vehicular Nets, etc.), and there is a growing interest from several research communities in doing inference over networks, performing resource allocation distributively, optimizing utility functions over graphs, diffusing information among agents, etc. The aim of this talk is to illustrate several recent advances in the field of graph signal processing (GSP) and distributed optimization over networks. In the first part of the talk, we introduce a novel algorithmic framework for adaptive graph signal processing, encompassing several different aspects. As a first step, we design sampling strategies and recovery algorithms for (distributed) reconstruction and tracking of dynamic graph signals observed over a selected set of (possibly time-varying) vertices. Then, we merge the sensing and communication aspects of the graph signal recovery problem, thus jointly optimizing transmission powers, quantization bits, and sampling set, while at the same time guaranteeing a target performance of the recovery task. Finally, we illustrate algorithms for learning a sparse graph topology to explain the relationships among some observed signals, with the aim of building an approximatively band-limited representation of the signals over the learned graph, and thus enable effective graph-based learning techniques. After illustrating the main tools, we will concentrate on specific applications such as, e.g., prediction of the traffic map in a city environment, cartography of the electromagnetic field from sporadic measurements, and inference of the functional connectivity of the brain. In the second part of the talk, we introduce a powerful algorithmic framework for solving nonconvex distributed optimization problems over graphs, where agents (a) have only local knowledge about the problem, and (b) exchange information according to a sparse (possibly time-varying) topology. The framework, termed as NEXT, builds on successive convex approximation techniques while using consensus as a mechanism for distributing the computations among the agents. NEXT represents the first algorithmic framework available in the literature with provable convergence to (stationary) solutions of general distributed nonconvex problems. Additionally, we show a principled way allowing each agent to exploit a possible multi-core architecture (e.g., a local cloud) in order to parallelize its local optimization step, resulting in strategies that are both distributed (across the agents) and parallel (inside each agent) in nature. The framework is very flexible and can be customized for several potential applications in signal processing, machine learning, communications, automatic control, etc. BIO: Paolo Di Lorenzo received the M.Sc. and the Ph.D. degrees in electrical engineering, both from Sapienza University, Rome, Italy, in 2008 and 2012, respectively. He is an Assistant Professor in the Department of Information Engineering, Electronics, and Telecommunications at Sapienza University. Previously, he was an Assistant Professor in the Department of Engineering at University of Perugia, Italy. In 2010, he held a visiting research appointment in the Department of Electrical Engineering, University of California at Los Angeles. He has participated in the European research project FREEDOM, on femtocell networks, SIMTISYS, on moving target detection through satellite constellations, and TROPIC, on distributed computing, storage and radio resource allocation over cooperative femtocells. He is currently involved in the H2020 European research project 5G-MiEdge, on multi-access edge computing enabled by millimeter wave communications. His current research interests include signal processing theory and methods, distributed optimization, adaptation and learning over networks, and graph signal processing. He received three Best Student Paper Awards, respectively, at the IEEE SPAWC’10, the EURASIP EUSIPCO’11, and the IEEE CAMSAP’11. He also received the 2012 GTTI Award for the Best Ph.D. thesis in information technologies and communications.

Permalink: https://www.eurecom/seminar/67101