The workshop will start with dinner on September 22 and end after lunch on September 25.

**Presentation slides**

**Monday September 23**

Rose Yu – Deep learning for spatiotemporal data

Ivan Markovsky – Data-driven structured noise filtering via common dynamics estimation

Tomas McKelvey – On recent advances in rank constrained optimization with applications to subspace based estimation methods

Michael Eichler – Causal Inference from Multivariate Time Series: Principles and Problems

Arun Venkitaraman – Graph Signal Processing and Machine Learning

Manfred Deistler (AUS) – On the sensitivity of Granger causality to errors-in-variables, linear transformations and subsampling

Xiaodong Cheng – Allocation of Excitation Signals for Generic Identifiability of Linear Dynamic Networks

**Tuesday September 24**

Donatello Materassi – Network recovery and identification via graphical separation tests

Tom Oomen – Numerically optimal identification of complex systems

**Wednesday Septemer 25**

Johan Karlsson – Optimal mass transport for tracking, estimation, and information fusion

Kévin Colin – Data informativity for the open-loop identification of MIMO systems in the Prediction Error framework

Michael Döhler – The local approach for vibration-based damage diagnosis of civil structures

Luca Zancato and Alessandro Chiuso – SGD tunneling through barrier potential exploiting the Loss Landscape of Neural Nets

Valentina Breschi and Simone Formetin – Direct data-driven design of switching controllers

**Plenary speakers**

#### Dr. Qi (Rose) Yu

**Affiliation**: Khoury College of Computer Sciences, Northeastern University

**Title**: Deep Learning For Spatiotemporal Data

**Abstract**: Applications such as climate science, intelligent transportation, aerospace control, and sports analytics apply machine learning for large-scale spatiotemporal data. This data is often nonlinear, high-dimensional, and demonstrates complex spatial and temporal correlations. Existing deep learning models cannot handle complex spatiotemporal dependency structures. We’ll explain how to design deep learning models to learn from large-scale spatiotemporal data, especially for dealing with non-Euclidean geometry, long-term dependencies, and logical and physical constraints. We’ll showcase the application of these models to problems such as long-term forecasting for transportation, long-range trajectories synthesis for sports analytics, and combating ground effect in quadcopter landing for aerospace control.

**Bio**: Dr. Yu is an Assistant Professor in the Khoury College of Computer Sciences at Northeastern University. Previously, she was a postdoctoral researcher in Caltech Computing and Mathematical Sciences. She earned her Ph.D. in Computer Sciences at the University of Southern California and was a visiting researcher at Stanford University. Her research focuses on developing machine learning techniques for large-scale time series and spatiotemporal data. She is generally interested in the theory and applications of deep learning, tensor optimization, and spatiotemporal modeling. Her work has been successfully applied to intelligent transportation, climate informatics, and aerospace control. Among her awards, she has won the best dissertation award in USC computer science, best paper award at NIPS time series workshop, and was nominated as one of the “MIT Rising Stars in EECS’.

#### Donatello Materassi

**Affiliation**: University of Minnesota, Twin Cities

**Title**: Network recovery and identification via graphical separation tests

**Abstract**: Networks have become ubiquitous in science. The principal advantages provided by a networked system are a modular approach to design, the possibility of directly introducing redundancy and the realization of distributed and parallel algorithms. All these advantages have led to consider networked systems in the realization of many technological devices. At the same time, it is not surprising that natural and biological systems tend to exhibit strong modularity, as well.

Interconnected systems are successfully exploited to perform novel modeling approaches in many fields, such as Economics, Biology, Cognitive Sciences, Ecology and Geology.

While networks of dynamical systems have been deeply studied and analyzed in physics and engineering, there is a reduced number of results addressing the problem of recovering and identifying an unknown network of dynamic systems, since it poses formidable theoretical and practical challenges. One of the main challenges is the identification of networked systems that are difficult to access or manipulate. Thus, the necessity for general tools for the identification of networks that are known only via non-invasive observations is rapidly emerging.

The talk addresses this problem leveraging approaches from the theory of graphical models and showing under what conditions they can be extended to networks of dynamic systems.

**Bio**: Donatello Materassi holds a Laurea in “Ingegneria Informatica” and a “Dottorato di Ricerca” in Nonlinear Dynamics and Complex Systems from Universita’ degli Studi di Firenze, Italy. He has been a research associate at University of Minnesota (Twin Cities) from 2008 till 2011. He has been concurrently both a post-doctoral researcher at Laboratory for Information and Decision Systems (LIDS) at the Massachusetts Institute of Technology and a lecturer at Harvard University till 2014.

He was an assistant professor at University of Tennessee in Knoxville from 2014 till 2018. Currently, he is an assistant professor at the University of Minnesota, Twin Cities. In 2016, he was a recipient of the NSF CAREER award. His main research interests are graphical models, stochastic systems and cybernetics.

#### Michael Eichler

**Affiliation**: University of Maastricht

**Title**: Causal Inference from Multivariate Time Series: Principles and Problems

**Abstract**: In time series analysis, inference about cause-effect relationships among multiple time series is commonly based on the concept of Granger causality, which exploits temporal structure to achieve causal ordering of dependent variables. One major and well known problem in the application of Granger causality for the identification of causal relationships is the possible presence of latent variables that affect the measured components and thus lead to so-called spurious causalities. This raises the question about whether Granger causality is an appropriate tool for causal learning; indeed, there are many researchers that deny any such claim.

To answer the question in more depth, we present a graph-theoretic approach for describing and analysing Granger-causal relationships in multivariate time series that are possibly affected by latent variables. It is based on mixed graphs in which directed edges represent direct influences among the variables while dashed edges – directed or undirected – indicate associations that are induced by latent variables. We show how such representations can be used for inductive causal learning from time series and discuss the underlying assumptions and their implications for causal learning. Finally we will discuss non-Markovian constraints imposed by latent variable structures and how these can be exploited for causal inference.

**Bio**: Michael Eichler is Associate Professor of Econometrics in the School of Business and Economics at Maastricht University. Previously, he was postdoctoral researcher at the Institute of Applied Mathematics of the University of Heidelberg, where he also earned his Ph.D. in Mathematics. He was also visiting assistant professor at the Department of Statistics of the University of Chicago and research associate at the Department of Statistical Science of University College London.

The main focus of his research is on causal modelling of multivariate time series and the development of algorithms for inferring causal networks. Further current research interests are functional time series and Bayesian methods for complex time-dependent systems.