Plenary Speakers
In addition to the embedded videos present in the description of each of the plenaries, the links to all the plenary talks are summarized next:
 Introduction to the GPS2020 workshop and plenary talks
 GSP2020PT1: Daniel P. Palomar (Learning Graphs of Stocks)
 GSP2020PT2: Edwin Hancock (Entropic Analysis of Network Time Series)
 GSP2020PT3: Patrick J. Wolfe (Modeling variation in network populations)
 GSP2020PT4: Sergio Barbarossa (Topological Signal Processing)

Daniel P. Palomar
Industrial Engineering & Decision Analytics
Hong Kong University of Science and Technology
Hong Kong, SAR 
Learning Graphs of Stocks: From iid to TimeVarying Models
Graphical representations of data are increasingly important tools used to uncover hidden relationships among variables. In stock markets, one is generally interested in learning about dependencies among stocks and how to leverage this information into practical scenarios such as clustering, portfolio design, and crisis analysis. While graphical models have been extremely successful across many fields including healthcare, bioinformatics, and social networks, they have yet to show their full potential in the financial markets. Financial data exhibit stylized facts that make modeling extremely challenging such as outliers, heavytailed distributions, and nonstationarity. In this talk, we will start by discussing several graph learning frameworks from a practitioner's perspective. Then, we will propose extensions tailored to financial data.
More info about the speaker here.

Edwin Hancock
Computer Science
University of York
UK 
Entropic Analysis of Network Time Series
Efficiently computing the entropy of a network has proved to be an elusive problem, with potentially enormous impact on the fields of machine learning, complex systems and big data analysis. In this talk I will present an overview of recent work that has shown how ideas from spectral graph theory and statistical physics can together be brought to bare on the problem, yielding simple and efficient methods for computing network entropy. With the entropy to hand, then a variety of principled information theoretic methods can be brought to bear to develop new tools for analysing network time series. Examples include information theoretic graph kernels, the minimum description length learning of generative network models, and entropy component analysis.
These new tools open up the possibility of addressing a number of practical problems including a) detecting anomalies in network time series, b) modelling the time evolution of networks and c) clustering different rtypes of network structure. I will furnish examples from the financial and medical domains to illustrate the application of the new tools in these domains.
More info about the speaker here.

Patrick J. Wolfe
Statistics & Computer Science
Purdue University
USA 
Modeling variation in network populations
How do we draw sound and defensible conclusions and accomplish signal processing tasks when working with populations of networks, for example in comparing two sets of network observations, or evaluating the appropriateness of new statistical network models? This talk will focus on areas of recent progress in understanding variability in network populations, along with the transformation of this understanding into new signal processing methods to model and draw inferences from network data in the real world. Along with practical examples and applications, the insights that result from connecting theory to practice also feed back into pure mathematics and theoretical computer science, prompting new questions at the interface of signal processing and these fields.
More info about the speaker here.

Sergio Barbarossa
Information Engineering, Electronics and Telecommunications
Sapienza University of Rome
Italy 
Topological Signal Processing: Uncovering patterns in the data relying on multiway relations
The goal in this talk is to establish the fundamental tools to analyze signals defined over a topological space, i.e. a set of points along with a set of neighborhood relations. The resulting framework can be defined as Topological Signal Processing (TSP). TSP does not necessarily require the introduction of a metric and it is then especially useful to deal with signals defined over nonmetric spaces. While the focus of the talk will be mostly on signals defined over simplicial complexes, extensions to more general structures, like hypergraphs, will be briefly discussed. The field of Graph Signal Processing (GSP) represents a simple case of TSP, as it refers to the situation where the observed signals are associated only with the vertices of a graph. Even though the theory can be applied to signals of any order, in the talk we focus on signals defined over the edges of a graph and show how building a simplicial complex of order two, i.e. including triangles, yields benefits in the analysis of edge signals. After reviewing the basic principles of algebraic topology, we derive a sampling theory for signals of any order and emphasize the interplay between signals of different order. Then we propose a method to infer the topology of a simplicial complex from data. We conclude with applications to real edge signals and to the analysis of a discrete vector field to illustrate the benefits of the proposed methodologies.
More info about the speaker here.