Abstract: In this tutorial lecture we present some key use-cases of challenging optimisation problems - centralised and distributed - in which it is shown that a control theoretic approach can have a considerable impact, especially when robustness is considered. Methodologies and tools such as Lyapunov analysis, ISS stability, small-gain theory, time-scales separation, and passivity are all instrumental to unveil fundamental properties of optimisation methods with a significant potential impact on key application areas such as multi-agent systems and large-scale critical infrastructures.

In the first use-case, we consider the discrete-time Arrow-Hurwicz-Uzawa primal-dual algorithm, also known as the first-order Lagrangian method, for constrained optimisation problems involving a smooth strongly convex cost and smooth convex constraints. We deal with the long-standing open problem of nonlocal asymptotic stability of such an algorithm. In particular, it is proved that an optimal equilibrium exists, it is unique, and it is semi-globally exponentially stable. We also show that, in the presence of constraints, global asymptotic stability cannot be established for the considered algorithm; hence, semi-global guarantees are the best achievable in general.

In the second use-case, we revisit the distributed version of the unconstrained Arrow-Hurwicz-Uzawa primal-dual algorithm, which was introduced in 2010 by J. Wang and N. Elia. Here, by means of a Lyapunov-based analysis, we prove global ISS of the algorithm relative to a closed invariant set composed of optimal equilibria and with respect to perturbations affecting the algorithm’s dynamics. In the absence of perturbations, this result implies linear convergence of the local estimates and Lyapunov stability of the optimal steady state. Moreover, we unveil fundamental connections with the well-known Gradient Tracking algorithm and with distributed integral control.

In the third use-case, as an extension of the Wang-Elia algorithm, we present a control approach to the problem of distributed optimisation for both continuous-time and discrete-time, linear uncertain multi-agent systems. Since all the agent dynamics is subject to parametric uncertainties and the gradient of the local objective function can only be measured through each agent’s output, the output agreement of multi-agents cannot be achieved by traditional distributed optimisation algorithms. Instead, we propose a simple internal-model-based tracking controller as a new module, integrated into the Wang-Elia algorithm to accomplish the output agreement with guaranteed exponential convergence.

Finally, early preliminary attempts to address the distributed version of the constrained discrete-time Arrow-Hurwicz-Uzawa primal-dual algorithm by a control theoretic approach are briefly sketched showing that passivity tools are instrumental in this challenging context.

Bio: Thomas Parisini received the Ph.D. degree in electronic engineering and computer science from the University of Genoa in 1993. He was with Politecnico di Milano. He currently holds the Chair of industrial control and the Head of the Control and Power Research Group, Imperial College London. He is the Deputy Director of the KIOS Research and Innovation Centre of Excellence, University of Cyprus. Since 2001, he has been the Danieli Endowed Chair of automation engineering with the University of Trieste. From 2009 to 2012, he was the Deputy Rector of the University of Trieste. In 2016, he was awarded as the Principal Investigator at Imperial College of the H2020 European Union Flagship Teaming Project, KIOS Research and Innovation Centre of Excellence, led by the University of Cyprus. In 2023, he received the Knighthood of the Order of Merit of the Italian Republic for Scientific Achievements Abroad; and the Honorary Doctorate from the University of Aalborg, Denmark, in 2018. He held a ”Scholar-in-Residence” visiting position with Digital Futures-KTH, Stockholm, Sweden. He has authored or coauthored a research monograph in the Communication and Control series (Springer and Nature); and over 400 research papers in archival journals, book chapters, and international conference proceedings. He is a fellow of IFAC. He was a co-recipient of the IFAC Best Application Paper Prize of the Journal of Process Control (Elsevier), from 2011 to 2013, for three years; and the 2004 Outstanding Paper Award of IEEE TRANSACTIONS ON NEURAL NETWORKS. He was a recipient of the 2007 IEEE Distinguished Member Award. He has served as the 2021–2022 President for the IEEE Control Systems Society. Among other activities, he was the Program Chair of the 2008 IEEE Conference on Decision and Control and the General Co-Chair of the 2013 IEEE Conference on Decision and Control. From 2009 to 2016, he was the Editor-in-Chief of IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY. Since 2017, he has been an Editor of Control Applications of Automatica and the Editor-in-Chief of the European Journal of Control, since 2018.

Abstract: Next generation of networked cyber-physical  systems will support a number of application domains e.g. connected autonomous vehicular networks, collaborative robotics in smart factories, and many other mission-critical applications. With the advent of massive machine-to-machine communication and IoT networks, huge volumes of data can be collected and processed with low latency through edge computing facilities.  Distributed machine learning  enables  cross-device  collaborative learning without exchanging raw data, ensuring privacy and reducing communication cost. Learning over wireless networks poses significant challenges due to limited communication bandwidth and channel variability, limited computational resources at the IoT devices, the heterogeneous nature of distributed data, and also randomly time-varying network topologies. In this talk, we will present  (i) low-complexity communication efficient Federated Learning (FL) algorithms based on approximate Newton-type optimization techniques employed at the local agents, which achieve superlinear convergence rate as opposed to linear rates achieved by state-of-the-art gradient descent based algorithms, and (ii) fully distributed network Newton type algorithms based on a distributed version of the well-known GIANT algorithm. While consensus based distributed optimization algorithms are naturally limited to linear convergence rates, we will show that one can design finite-time consensus based distributed network-Newton type algorithms that can achieve superlinear convergence with respect to the number of Newton steps, albeit at the cost of increased numbers of consensus rounds. We will conclude with some recent results on zeroth order Hessian approximation techniques that can also achieve superlinear convergence rates in Federated Learning.

Slides 

Bio: Subhrakanti Dey received the Ph.D. degree from the Department of Systems Engineering, Research School of Information Sciences and Engineering, Australian National University, Canberra, in 1996. He is currently a Professor and Head of the Signals and Systems division in the Dept of Electrical Engineering at Uppsala University, Sweden. He has also held professorial positions at NUI Maynooth, Ireland and University of Melbourne, Australia. His current research interests include networked control systems, distributed machine learning and optimization, and detection and estimation theory for wireless sensor networks. He is a Senior Editor for IEEE Transactions of Control of Network Systems and IEEE Control Systems Letters, and an Associate Editor for Automatica. He is a Fellow of the IEEE.

Abstract: In today's interconnected world, attention is drawn towards a new generation of intelligent (mobile) cooperative autonomous systems, built with increasingly powerful sensors that allow them to cooperate with other autonomous systems. The cooperation of autonomous systems, which is an emerging technology in many areas (e.g., intelligent transportation systems and factory automation) is inevitably done over a wireless network.  Effective communication plays a pivotal role in the successful operation of distributed systems and optimization processes. This work delves into the realm of communication-aware distributed coordination and optimization, exploring how wireless communication could be implemented in coordination strategies to enhance efficiency, scalability, and robustness in such distributed systems. 

Slides 

Bio: Themistoklis Charalambous received his BA (First Class Honours) and M.Eng (Distinction) in Electrical and Information Sciences from Trinity College, Cambridge University. He completed his PhD studies in the Control Laboratory, of the Engineering Department, Cambridge University in 2009. Following his PhD, he joined the Human Robotics Group as a Research Associate at Imperial College London for an academic year (September 2009-September 2010). Between September 2010 and December 2011, he worked as a Visiting Lecturer at the Department of Electrical and Computer Engineering, University of Cyprus. Between January 2012 and January 2015, he worked at the Division of Decision and Control of the Department of Intelligent Systems at the Royal Institute of Technology (KTH) as a Postdoctoral Researcher. Between April 2015 and December 2016, he worked as a Postdoctoral Researcher in the Unit of Communication Systems at the Department of Electrical Engineering of Chalmers University of Technology. In January 2017, he joined Department of Electrical Engineering and Automation, School of Electrical Engineering, Aalto University as a tenure-track Assistant Professor. In September 2018, he was awarded the Academy of Finland Research Fellowship and in July 2020 he was appointed as a tenured Associate Professor. In September 2021, he joined the Department of Electrical and Computer Engineering, University of Cyprus as a tenure-track Assistant Professor and he remains associated with Aalto University as a Visiting Professor. Since April 2023, he is also a Visiting Professor at the FinEst Centre for Smart Cities. His primary research targets the design and analysis of (wireless) networked control systems that are stable, scalable and energy efficient. The study of such systems involves the interaction between dynamical systems, their communication and the integration of these concepts. As a result, his research is interdisciplinary combining theory and applications from control theory, communications, network and distributed optimization.

Abstract: In this talk I will address control and learning scenarios in which models of complex systems are not fully available. This results in optimization and optimal control (reinforcement learning) problems in which the cost function and/or the constraints are not completely known. I will show key challenges arising in addressing these problems as, e.g., large scale nature of the systems, structure of the required policy, on-policy and closed-loop nature of the learning strategy, local communication requirements. Energy and robotic systems are key sources of concrete scenarios in which these challenges arise. New strategies addressing the above challenges will be shown with applications to these key domains and future perspectives.

Bio: Giuseppe Notarstefano is a Professor in the Department of Electrical, Electronic, and Information Engineering G. Marconi at Alma Mater Studiorum Università di Bologna. He was Associate Professor (from June ‘16 to June ‘18) and previously Assistant Professor, Ricercatore, (from February ‘07) at the Università del Salento, Lecce, Italy. He received the Laurea degree “summa cum laude” in Electronics Engineering from the Università di Pisa in 2003 and the Ph.D. degree in Automation and Operation Research from the Università di Padova in 2007. He has been visiting scholar at the University of Stuttgart, University of California Santa Barbara and University of Colorado Boulder. His research interests include distributed optimization, cooperative control in complex networks, applied nonlinear optimal control, and trajectory optimization and maneuvering of aerial and car vehicles. He serves as an Associate Editor for IEEE Transactions on Automatic Control, IEEE Transactions on Control Systems Technology and IEEE Control Systems Letters. He is also part of the Conference Editorial Board of IEEE CSS and EUCA. He is recipient of an ERC Starting Grant 2014.

Abstract: Networked Control Systems have been the subject of extensive research over the last two decades. However, their development is still refrained through multifold design challenges. Indeed, interconnected subsystems make it hard to design local decision-making, whereas global design is typically undesired if not infeasible. The lack of formal guarantees is replaced in practice with heuristics built on human intuition and experience: a cornerstone rule of thumb is that allocating more resources leads to higher performance, or, broadly speaking, that “more is better”. In this talk, I will dig into, and possibly question, the role that such a common belief plays in the design of Networked Control Systems. I will consider three common design assumptions: (i) more sensors improve estimation quality; (ii) more communication links increase control performance; (iii) more collaboration enhances cooperative tasks. While these assumptions are reasonable in general, I will discuss three scenarios where they fail to capture the true nature of the system: (i) more sensors may hinder estimation under computational delays; (ii) more communication links may degrade control performance under communication constraints; (iii) more collaboration may be harmful in the presence of misbehaving agents. These considerations urge us to carefully (re-)think suitable design strategies, which are preliminarily investigated in the talk.

Bio: Luca Schenato received the Dr. Eng. degree in electrical engineering from the University of Padova in 1999 and the Ph.D. degree in Electrical Engineering and Computer Sciences from the UC Berkeley, in 2003. He held a post-doctoral position in 2004 and a visiting professor position in 2013-2014 at U.C. Berkeley. Currently he is Full Professor at the Information Engineering Department at the University of Padova. His interests include networked control systems, multi-agent systems, wireless sensor networks, distributed optimisation and synthetic biology. Luca Schenato has been awarded the 2004 Researchers Mobility Fellowship by the Italian Ministry of Education, University and Research (MIUR), the 2006 Eli Jury Award in U.C. Berkeley and the EUCA European Control Award in 2014, and IEEE Fellow in 2017. He served as Associate Editor for IEEE Trans. on Automatic Control from 2010 to 2014 and he is he is currently Senior Editor for IEEE Trans. on Control of Network Systems, Associate Editor for Automatica and Chair of IFAC TC on Networked Systems. 

Abstract: In this talk, we will first review the state of the art on equilibrium seeking in multi-agent systems with focus on complex features such as incomplete and sparse information, stochasticity and existence of multiple equilibria. Next, we will introduce a novel receding-horizon framework for equilibrium seeking in real time. Finally, we will outline several open research directions.

Slides 

Bio: Sergio Grammatico is an Associate Professor at the Delft Center for Systems and Control, TU Delft, The Netherlands. He received the Bachelor’s degree in Computer Engineering, the Master’s degree in Automatic Control, and the Ph.D. degree in Automatic Control, all from the University of Pisa, Italy, in 2008, 2009, and 2013 respectively. In 2013–2015, he was a postdoc researcher in the Automatic Control Laboratory, ETH Zurich, Switzerland. In 2015–2018, he was an Assistant Professor in the Department of Electrical Engineering, Control Systems, TU Eindhoven. He was a recipient of the Best Paper Award at the 2016 ISDG Int. Conf. on Network Games, Control and Optimization, of the 2021 Roberto Tempo Best CDC Paper Award, and co-author for the 2022 IEEE CSS Italy Young Author Best Journal Paper Award. He is currently an Associate Editor of the IEEE Trans. on Automatic Control and of IFAC Automatica. His research interests include dynamic game theory, multi-agent systems and extremum seeking control.