Abstracts

• Speaker: Ari Arapostathis
  • Title: On the relative value iteration for discrete and continuous time Markovian models.
  • Abstract: Abstract: We present various results concerning the convergence of the relative value iteration algorithm in a unified framework. These include results for controlled diffusions under a coercive penalty function, and also controlled Markov processes with continuous state space. We also demonstrate how the rate of convergence to the stationary distribution of the process under the optimal control affects the convergence of the relative value iteration, by analyzing the special case when the process under the optimal control is geometrically ergodic. The predominant part of the talk is based on joint work with Vivek S. Borkar and K. Suresh Kumar.
  • Bio: Ari Arapostathis is a professor at the Department of Electrical and Computer Engineering at the University of Texas at Austin. His research interests include analysis and estimation techniques for stochastic systems, stability properties of large-scale interconnected power systems, and stochastic and adaptive control theory. His main technical contributions are in the areas of adaptive control and estimation of stochastic systems with partial observations, controlled diffusions, adaptive control of nonlinear systems, geometric nonlinear theory, and stability of large scale interconnected power systems. His research is currently funded by the National Science Foundation (Division of Mathematical Sciences), Army Research Office (Applied Probability), and the Office of Naval Research. He is a Fellow of IEEE.

Slides are embedded below:

• Speaker: Marco Campi
  • Title: Data-driven Decision Making for Control, Identification and Classification: the Scenario Approach
  • Abstract: In the present era where data are increasingly gaining prominence, there is a need for methodologies able to merge domain knowledge with information conveyed by data. The scenario optimization is a general approach that enables one to make data-driven designs. When the scenario design is applied to a new, out-of-sample, case, its performance is guaranteed by a solid generalization theory which underpins the use of the method. In this talk, I shall try to give an overview of the scenario approach, including its theoretical foundations. The generality of the scenario approach makes it useful across a variety of fields including predictive control, identification and classification and examples will be provided to highlight the versatility of the method.
  • Bio: Marco Claudio Campi is professor of control and inductive methods at the University of Brescia, Italy. He has been in various capacities on the Editorial Board of Automatica, Systems and Control Letters and the European Journal of Control, and he was until 2017 the chair of the Technical Committee IFAC on Modeling, Identification and Signal Processing (MISP). Marco Campi is a recipient of the "Giorgio Quazza" prize, and he received in 2008 the IEEE CSS George S. Axelby outstanding award for the article "The Scenario Approach to Robust Control Design". He has delivered plenary and semi-plenary addresses at major conferences including CDC, SYSID, ECC, and MTNS. Currently he is a distinguished lecturer of the Control Systems Society. Marco Campi is a Fellow of IEEE, a member of IFAC, and a member of SIDRA.

Slides and a course announcement are embedded below:

• Speaker: Peyman Mohajerin Esfahani
  • Title: Data-driven Decision-Making: A Distributionally Robust Approach
  • Abstract:We consider decision-making problems where the distribution of the uncertain parameters is only observable through a finite training dataset. After a short overview of the existing techniques toward such problems, we focus on a modeling paradigm termed as distributionally robust optimization (DRO) where the objective is to find a decision that minimizes the worst-case expected cost over a family of distributions characterized through certain known properties of the unknown data-generating distribution. In this talk, we construct the family of distributions as a Wasserstein ball in the space of (multivariate and non-discrete) probability distributions centered at the uniform distribution on the training samples. We study the computationally tractability of the resulting optimization programs, the finite-sample guarantee of the outcome decisions, and also a hidden connection to the well-known regularization techniques in the machine learning literature. Furthermore, if time permits, we will also discuss important problems of inverse optimization, maximum likelihood estimation, and Kalman filtering through a DRO lens.
  • Bio: Peyman Mohajerin Esfahani is an assistant professor in the Delft Center for Systems and Control at the Delft University of Technology. Prior to joining TU Delft, he held several research appointments at EPFL, ETH Zurich, and MIT between 2014 and 2016. He received the B.Sc. and M.Sc. degrees from Sharif University of Technology, Iran, and the PhD degree from ETH Zurich. His research interests include theoretical and practical aspects of decision-making problems in uncertain and dynamic environments, with applications to control and security of large-scale and distributed systems. He was selected for the Spark Award by ETH Zurich for the twenty best inventions of the year in 2013, and was one of the three finalists for the Young Researcher Prize in Continuous Optimization awarded by the Mathematical Optimization Society in 2016. He was a recipient of the 2016 George S. Axelby Outstanding Paper Award from the IEEE Control Systems Society, an award that recognizes the best paper published in the past two years in the IEEE Transactions on Automatic Control.

Slides are embedded below:

• Speaker: Rolf Findeisen
  • Title: Learning supported predictive control for autonomous systems with guarantees
  • Abstract: Predictive control approaches are increasingly used for the control and operation of autonomous systems, spanning from robotics to autonomous driving and the operation of chemical and biotechnological processes. This success is based on the possibility to use process models, as well as to take constraints and preview information’s of disturbances and set point changes directly into account. However, large uncertainties, disturbances, as well as uncertainties in the model of the system and environment render the application of predictive control challenging. Fusing predictive control approaches with learning approaches, either to adapt – learn - suitable process models or to learn the control law itself, taking measurement data into account, is very promising to overcome the appearing challenges. In the frame of this contribution we provide an introduction to the fusion of machine learning approaches and predictive control. We focus on the question how guarantees despite learning can be provided. We specifically outline how the combination of set-based approaches with learning strategies in the case of systems that show multiple operational modes, as well predictive control methods using models which include Gaussian model components can provide guarantees. The results are underlined by several examples from robotics to chemical and biotechnological processes..
  • Bio: Rolf Findeisen obtained a M.S. degree from the University of Wisconsin, Madison, a Diploma in Engineering Cybernetics and Doctorate from the University of Stuttgart. Since 2007 he is heading the Laboratory for Systems Theory and Automatic Control at the Otto-von-Guericke Universität Magdeburg. Rolf is editor/associated editor of several journals including the IEEE Transactions on Control of Network Systems and he is the international program co-chair of the IFAC World Congress 2020 in Berlin. The research of his group focuses on optimal and predictive control, control for autonomous systems, learning and control, decision-making under uncertainty, and network controlled systems. The considered fields of applications span from mechatronics, robotics, biotechnology, chemical processes, up to medicine.

Slides are embedded below:

• Speaker: Lars Grüne
  • Title: Infinite horizon performance estimates for Nonlinear Model Predictive Control
  • Abstract: The talk will discuss different settings under which rigorous infinite horizon performance estimates for the NMPC closed loop can be established. We will discuss stabilizing and economic MPC, schemes with and without terminal conditions and discounted and undiscounted problems. We will also discuss theoretical foundations regarding strict dissipativity and the turnpike property, two properties which are pivotal for the analysis. The results will be for deterministic MPC schemes but an outlook including some conjectures for stochastic schemes will be given.
  • Bio: Lars Grüne has been Professor for Applied Mathematics at the University of Bayreuth, Germany, since 2002. He received his Diploma and Ph.D. in Mathematics in 1994 and 1996, respectively, from the University of Augsburg and his habilitation from the J.W. Goethe University in Frankfurt/M in 2001. He held visiting positions at the Universities of Rome ‘Sapienza’ (Italy), Padova (Italy), Melbourne (Australia), Paris IX — Dauphine (France) and Newcastle (Australia). Prof. Grüne is Editor-in-Chief of the journal Mathematics of Control, Signals and Systems (MCSS) and Associate Editor of several other journals, including the Journal of Optimization Theory and Applications (JOTA) and the IEEE Control Systems Letters. His research interests lie in the area of mathematical systems and control theory with a focus on numerical and optimization-based methods for nonlinear systems.

Slides are embedded below:

• Speaker: Colin Jones
  • Title: Optimal Synthesis of Fixed-Structure Embedded Optimization-Based Controllers
  • Abstract: The field of fast-MPC, or the use of embedded optimization for high speed control, is a rapidly growing field in academia and increasingly in industry. Achieving the required extremely high speed optimization, often within micro-seconds, on low-end embedded platforms calls for a wide range of heuristic procedures for both the control design, as well as in the implementation of the optimization algorithms themselves. This semi-heuristic process leads to complex control laws that can be very efficient, but that are also extremely difficult to tune and design. This talk will introduce a framework for the non-conservative analysis of many of the heuristics used in these controllers via a convex sum-of-squares approach. We will then build on this framework to develop a formal optimal synthesis procedure for very high-speed embedded optimization-based control laws, and give a number of examples.
  • Bio: Colin Jones has been an Associate Professor in the Automatic Control Laboratory at the EPFL in Switzerland since 2017. He was a Senior Researcher at the Automatic Control Lab at ETH Zurich until 2010 and obtained a Ph.D. in 2005 from the University of Cambridge for his work on polyhedral computational methods for constrained control. Prior to that, he was at the University of British Columbia in Canada, where he took his bachelor and master degrees in Electrical Engineering and Mathematics. His current research interests are in the areas of high-speed predictive control and optimization, as well as green energy generation, distribution and management.

Slides are embedded below:

• Speaker: Jean-Bernard Lasserre
  • Title: The Moment-SOS Hierarchy and some of its Applications
  • Abstract: The Moment-SOS Hierarchy was initially designed to compute the global minimum of polynomial optimization problems. Its convergence to the global minimum relies on powerful positivity certificates from Real Algebraic Geometry in the nineties. In fact the same methodology also applies for solving the ``Generalized Moment Problem" (GMP) with algebraic data, whose list of potential important applications in various fields is almost endless. We will describe some of them, including polynomial optimization, control, optimal control and inverse optimal control problems, a class of hyperbolic PDEs, and linear optimization problems on measures (Super resolution, optimal design, computational geometry, etc.)
  • Bio: Graduated from "Ecole Nationale Supérieure d'Informatique et Mathematiques Appliquees" (ENSIMAG) in Grenoble, then got his PhD (1978) and "Doctorat d'Etat"(1984) degrees both from Paul Sabatier University in Toulouse (France). He has been at LAAS-CNRS in Toulouse since 1980, where he is currently Directeur de Recherche. He is also a member of IMT, the Institute of Mathematics of Toulouse. Twice a one-year visitor (1978-79 and 1985-86) at the Electrical Engineering Department. of UC Berkeley with a fellowship from INRIA and NSF. Several visits to Stanford University, MIT, UC Berkeley and MSRI, the Fields Institute, Toronto, IMA (Minneapolis), IPAM (UCLA), Cinvestav-IPN (Mexico), Leiden University and the Tinbergen Institute (Amsterdam, The Netherlands), the University of Adelaide and the University of South Australia (Adelaide), the University of New South wales (Sydney), the University of British Columbia (Vancouver).

Slides are embedded below:

• Speaker: Daniel Limon
  • Title: Model predictive control for changing operating conditions
  • Abstract: Nowadays, Model Predictive Control is considered a mature control technique with well-established design conditions to ensure stability and constraint satisfaction. However, these properties may be lost is the operating conditions of the control system are changed. This might be induced from unexpected changes in the references or in the disturbances or variations on the unitary costs in the economic optimization module.This talk is devoted to present recent results on the design of predictive controllers capable to cope with these changes ensuring closed-loop stability. This will be addressed from both, a tracking and an economic control point of view. Besides steady state as well as periodic operation conditions will be studied.
  • Bio: Daniel Limon is Professor at the Department of Systems Engineering and Automation of the University of Seville (Spain) and he is the responsible of the research group on Estimation, Prediction, Optimization and Control. He is the author or coauthor of more than 100 publications including book chapters, journal papers, conference proceedings and educational books. He was Keynote Speaker at the International Workshop on Assessment and Future Directions of Nonlinear Model Predictive Control in 2008 and Semiplenary Lecturer at the IFAC Conference on Nonlinear Model Predictive Control in 2012 (NMPC’12). He has been the Chair of the 5th IFAC Conference on Nonlinear Model Predictive Control (NMPC’15). His current research interests include Model Predictive Control, economic process control, trajectory tracking control and data-driven control.

Slides are embedded below:

• Speaker: John Lygeros
  • Title: Approximate dynamic programming through finite dimensional linear programs
  • Abstract: It is known that many optimal control problems encoded as dynamic programs can equivalently be characterised through the solution of a linear program (LP). For systems with continuous state and action spaces, the resulting LP involves an infinite number of decision variables (e.g. taking values in the space of real valued functions of the state) and an infinite number of constraints (e.g. one inequality constraint for each state-action pair). Replacing the infinite LP by a finite dimensional counterpart provides a method for approximating the solution of the original optimal control problem, in the spirit of Approximate Dynamic Programming. The question is how good the resulting approximation is. We show how recent results on optimal value approximation in randomised optimisation can be leveraged to derive probabilistic bounds on the value function error incurred in such LP-based ADP approximations. As an initial step in the direction of data driven control, we also discuss how (at least in the case of deterministic systems) the approximation can be carried out using sample paths of the system state evolution, bypassing the need for identifying a model. The talk is based on joint work with Peyman Mohajerin, Tobias Sutter, Angeliki Kamoutsi, and Alexandros Tanzanakis.
  • Bio: John Lygeros completed a B.Eng. degree in electrical engineering in 1990 and an M.Sc. degree in Systems Control in 1991, both at Imperial College of Science Technology and Medicine, London, U.K.. In 1996 he obtained a Ph.D. degree from the Electrical Engineering and Computer Sciences Department, University of California, Berkeley. During the period 1996-2000 he held a series of research appointments at the National Automated Highway Systems Consortium, Berkeley, the Laboratory for Computer Science, M.I.T., and the Electrical Engineering and Computer Sciences Department at U.C. Berkeley. Between 2000 and 2003 he was a University Lecturer at the Department of Engineering, University of Cambridge, U.K., and a Fellow of Churchill College. Between 2003 and 2006 he was an Assistant Professor at the Department of Electrical and Computer Engineering, University of Patras, Greece. In July 2006 he joined the Automatic Control Laboratory at ETH Zurich, first as an Associate Professor, and since January 2010 as a Full Professor; he is currently serving as the Head of the laboratory. His research interests include modelling, analysis, and control of hierarchical, hybrid, and stochastic systems, with applications to biochemical networks, automated highway systems, air traffic management, energy systems, and camera networks. John Lygeros is a Fellow of the IEEE, and a member of the IET and the Technical Chamber of Greece; since 2013 he is serving as the Treasurer of the International Federation of Automatic Control.

Slides are embedded below:

• Speaker: Coorous Mohtadi
  • Title: Real-Time Optimisation in Control: From humble beginning to intelligent autonomous systems
  • Abstract: Design of control systems has always been revolving around multi-objective optimisations subject to real constraints. Model-Based Predictive Control concepts were from their early stages attempting to formalise such problem formulations. Successes in petrochemical industries combined with renewed research in universities helped the field to flourish earlier in this century. This subject is now common place in many degree courses. The real more innovative applications have come from applying the principles to such areas as intelligent and autonomous systems. By way of an example, ADAS and autonomous driving technologies are redefining the automotive industry, changing all aspects of transportation, from daily commutes to long-haul trucking. Engineers across the industry use Model-Based Design with MATLAB® and Simulink® to develop their automated driving systems. We will demonstrate how software environments such as MATLAB and Simulink can serve as an integrated development environment for the different domains required for automated driving, including perception, sensor fusion, and control design.
  • Bio: Coorous Mohtadi is a senior member of the MathWorks technical specialist team supporting universities focusing on the application of MATLAB and Simulink in laboratories and curriculum development. He is interested in finding synergies between industry and academia. He has been supporting design and development in industry and universities for the last 9 years at MathWorks. Prior to joining MathWorks in 2007, Coorous was the European technical manager for temperature, process control, and component products at Omron Electronics Europe and the chief control engineer at Eurotherm Controls. During 1980s he was also a postdoctoral research fellow at University of Oxford, U.K. and University of Alberta, Canada. Coorous holds a D.Phil. in model-based predictive control and Masters in engineering science, both from University of Oxford. His paper on generalised predictive control has over 3500 citations and his industrial algorithms forms the core of extremely successful Eurotherm Controls 2000 series of products.

Slides are embedded below:

• Speaker: M. Vidyasagar
  • Title: An Overview of Compressed Sensing
  • Abstract: Compressed sensing refers to the recovery of high-dimensional but low-complexity objects from a limited number of measurements. Examples of such objects include: high-dimensional but sparse vectors, where the locations of the nonzero components are unknown (known as vector recovery); and high-dimensional but low rank matrices where an upper bound on the rank is available, but that is all (known as matrix recovery). During the past fifteen years or so, there have been significant advances on both of these problems. Initial advances were based on probabilistic methods, wherein the unknown vector or matrix was projected onto several random subspaces. However, in recent times, the speaker and his students have been propagating the idea that generating the measurements in a deterministic fashion is more efficient. The resulting theory is a fascinating mixture of concepts from probability theory, graph theory, and optimization. In this talk, an overview will be given of some of the key results. The application of these results will be illustrated through numerical examples.
  • Bio: Mathukumalli Vidyasagar was born in Guntur, India on September 29, 1947. He received the B.S., M.S. and Ph.D. degrees in electrical engineering from the University of Wisconsin in Madison, in 1965, 1967 and 1969 respectively. Between 1969 and 1989, he was a Professor of Electrical Engineering at Marquette University, Concordia University, and the University of Waterloo. In 1989 he returned to India as the Director of the newly created Centre for Artificial Intelligence and Robotics (CAIR) in Bangalore, under the Ministry of Defence, Government of India. Between 1989 and 2000, he built up CAIR into a leading research laboratory with about 40 scientists and a total of about 85 persons, working in areas such as flight control, robotics, neural networks, and image processing. In 2000 he moved to the Indian private sector as an Executive Vice President of India's largest software company, Tata Consultancy Services. In the city of Hyderabad, he created the Advanced Technology Center, an industrial R&D laboratory of around 80 engineers, working in areas such as computational biology, quantitative finance, e-security, identity management, and open source software to support Indian languages. In 2009 he retired from TCS and joined the Erik Jonsson School of Engineering & Computer Science at the University of Texas at Dallas, as a Cecil & Ida Green Chair in Systems Biology Science. In March 2010 he was named as the Founding Head of the newly created Bioengineering Department, a position that he relinquished in July 2013. In January 2015 he received the Jawaharlal Nehru Science Fellowship and since then he has been dividing his time between UT Dallas and the Indian Institute of Technology Hyderabad. His research interests are in the broad area of system and control theory, and its applications. At present he is interested in the area of compressed sensing, that is, finding sparse solutions to large under-determined problems, and the intersection between compressed sensing and control theory. On the applications front, he is interested in applying ideas from machine learning to problems in computational biology with emphasis on cancer. Vidyasagar has received a number of awards in recognition of his research contributions, including Fellowship in The Royal Society, the world's oldest scientific academy in continuous existence, the IEEE Control Systems (Technical Field) Award, the Rufus Oldenburger Medal of ASME, the John R. Ragazzini Education Award from AACC, and others. He is the author of eleven books and more than 140 papers in peer-reviewed journals.

Slides are embedded below:

• Speaker: Serdar Yuksel
  • Title: Robustness to incorrect models and application to quantized approximations in stochastic control
  • Abstract: In stochastic control, typically, an ideal model is assumed or an estimate model is learned, and the control design is based on this model, raising the problem of performance loss due to the mismatch between the assumed model and the actual model. In addition, even when a complete model is available, often computational methods dictate the use of approximate models. We will first consider robustness to quantized approximations for stochastic control problems with standard Borel spaces and present conditions under which finite models obtained through quantization of the state and action sets can be used to construct approximately optimal policies. We do not impose restrictive regularity conditions such as Lipschitz continuity, often assumed in the literature. In particular, only weak continuity of the transition kernel in the state and action is sufficient for the convergence of finite approximations. Under further conditions, we obtain explicit rates of convergence to the optimal cost of the original problem as the quantization rate increases. We will also extend our analysis to decentralized stochastic control where one can construct a sequence of finite models whose solutions constructively converge to the optimal cost. We then investigate robustness to more general modeling errors and study the mismatch loss of optimal control policies designed for incorrect models applied to the true system, as the incorrect model approaches the true model under various criteria. We show that continuity and robustness cannot be established under weak and setwise convergences of transition kernels in general, but that the expected induced cost is robust under total variation. For partially observed models, by imposing further assumptions on the measurement channel, we show that the optimal cost can be made continuous under weak convergence of transition kernels as well. These entail implications on empirical learning in (data-driven) stochastic control since often system models are learned through empirical training data where typically weak convergence criterion applies but stronger convergence criteria do not. (Collaborators: Ali Kara, Naci Saldi and Tamas Linder).
  • Bio: S. Yuksel received his B.Sc. degree in Electrical and Electronics Engineering from Bilkent University; and M.S. and Ph.D. degrees in Electrical and Computer Engineering from the University of Illinois at Urbana-Champaign in 2003 and 2006. He was a post-doctoral researcher at Yale University before joining Queen's University in the Department of Mathematics and Statistics, where he is now an Associate Professor. He is a co-author of the books Stochastic Networked Control Systems, and Finite Approximations in Discrete-Time Stochastic Control, both published by Springer. He has been awarded the 2013 CAIMS/PIMS Early Career Award in Applied Mathematics in Canada. His research interests are on stochastic control, decentralized control, information theory, and probability. He is an Associate Editor for the IEEE Trans. on Automatic Control and Automatica.

Slides are embedded below:

25-29 Nov 2018, IIT Bombay