2026 International Workshop
on Sparse and Geometric Representations
in Systems and Control
(A Satellite Workshop prior to IFAC WC 2026) 

Abstract 

Energy-based learning is a biologically plausible alternative to the popular back-propagation method for training artificial neural networks. It deals with models that are governed by a parameterized energy function. Learning is achieved by shaping the energy function such that its minima coincide with given data. The energy-based perspective is a promising avenue for training analog circuits in an energy-efficient manner, especially because training can be performed using relatively simple, local parameter updates. The purpose of this talk is to make some steps towards a theoretical understanding of energy-based learning, applied to nonlinear resistive networks. For these networks, we propose an energy-based learning algorithm and we establish conditions under which this algorithm converges to a suitable vector of parameters, explaining the data. 

Bio 

Henk van Waarde is an assistant professor in the Bernoulli Institute for Mathematics, Computer Science and Artificial Intelligence at the University of Groningen in The Netherlands. During 2020-2021 he was a postdoctoral researcher, first at the University of Cambridge, UK, and later at ETH Zürich, Switzerland. He obtained the Ph.D. degree (cum laude) in Applied Mathematics from the University of Groningen in 2020. He was also a visiting researcher at the University of Washington, Seattle in 2019-2020. His research interests include learning and data-driven control, system identification and identifiability, networks of dynamical systems, and robust and optimal control. Dr. van Waarde is the recipient of the 2025 SIAM Activity Group on Control and Systems Theory Prize. He serves as an Associate Editor of the IEEE Control Systems Letters.

Abstract 

While numerous safety filter methods have been developed to ensure the integrity of autonomous systems, most existing frameworks rely heavily on precise prior knowledge of system dynamics and predefined safety constraints. Such assumptions are often violated in unpredictable outdoor applications, making practical deployment challenging. Furthermore, even when applicable, existing methods frequently suffer from overly conservative safety interventions that hinder the system's ability to achieve desirable behaviors. In this talk, we discuss the fundamental challenges of deploying safety-critical control in unstructured outdoor environments and introduce a safety filter framework designed to reduce unnecessary conservativeness while accommodating unknown dynamic structures. This approach seeks to maintain a conceptual safety envelope through adaptive mechanisms, allowing for high-performance maneuvers even in the presence of environmental uncertainty. Through a combination of theoretical developments and experimental results, this presentation demonstrates how the proposed framework enhances operational reliability and safety-awareness, effectively bridging the gap between formal safety theory and complex real-world applications.

Bio 

Donggun Lee, Ph.D., is an Assistant Professor in the Department of Mechanical and Aerospace Engineering at North Carolina State University. He earned his Ph.D. in Mechanical Engineering from the University of California, Berkeley, and subsequently served as a Postdoctoral Associate at the Massachusetts Institute of Technology (MIT). He is the director of the Intelligent Controls Lab (ICL) at NCSU, where his research group focuses on the intersection of control theory, formal methods, and robotics. Dr. Lee’s work is dedicated to developing scalable, provably safe control frameworks—specifically leveraging Hamilton-Jacobi reachability analysis—to enable autonomous systems to operate reliably in complex and uncertain outdoor environments.

Abstract 

This talk explores the hierarchy among different types of LTI system representations, in the sense that a single, higher-level representation branches into non-unique models at a lower level. This provides insights into how a “data-driven representation,” formulated from input-output data trajectories based on Willems’ fundamental lemma, covers multiple ARX forms or state-space realizations with different stability properties. In this regard, a method to specify certain models in lower-level representations from input-output data is proposed, in a way that benefits follow-up data-driven controller synthesis.

Bio 

Joowon Lee received the B.S. and combined M.S./Ph.D. degrees in electrical and computer engineering from Seoul National University, Seoul, South Korea, in 2019 and 2026, respectively. She is currently a Postdoctoral Researcher with Department of Decision and Control Systems, KTH Royal Institute of Technology, Stockholm, Sweden. Her research interests include data-driven control and encrypted control.

Abstract 

Hamilton-Jacobi reachability analysis provides us with a powerful framework for safe control when the dynamics of a system are known. However, when the dynamics are uncertain, it becomes more challenging to guarantee safety. In this presentation, we introduce the concept of the data-driven Hamiltonian (DDH), which circumvents the need for an explicit dynamics model in the Hamilton-Jacobi reachability computation. Our approach relies on (i) mild prior knowledge about the system (e.g., Lipschitz constants of the dynamics) and (ii) a dataset of observations of the system's behavior. This framework enables the construction of reachable sets directly from data. Our formulation ensures a conservative approximation of the true Hamiltonian, thus guaranteeing our estimated safe set is a conservative estimate of the true safe set. We also propose a data-efficient safe experiment framework to gradually expand our estimated safe set while ensuring that safety constraints are not violated during data collection. This is achieved by iteratively conducting experiments within the computed data-driven safe set and updating the set using newly collected trajectory data. To demonstrate the capabilities of our approach, we showcase its effectiveness in safe flight envelope expansion for a tiltrotor vehicle transitioning from near-hover to forward flight. We will also discuss ongoing work extending this framework to more general optimal control problems.  

Bio 

Chris is a PhD student in the Hybrid Systems Laboratory at UC Berkeley. He is interested in data-driven safe control and also studies how information is represented in the brain during driving. He is also interested in the application of control theory to bio-medical systems.

Abstract 

We develop harmonic-analytic foundations for the study of invariant positive-definite kernels on Riemannian symmetric spaces with applications including kernel construction on spaces of covariance matrices. New results provide practical criteria for positive-definiteness and universality in both compact and non-compact settings. Several examples and numerical experiments illustrate the theory and its applications to statistical learning problems.

Bio 

Cyrus is a National Research Foundation (NRF) Fellow and Assistant Professor in Applied Mathematics at the School of Physical and Mathematical Sciences at Nanyang Technological University (NTU) in Singapore. Cyrus is also a Bye-Fellow at Fitzwilliam College in the University of Cambridge, where he previously held an Early Career Research Fellowship from 2018 to 2022. He studied mathematics as an undergraduate at the University of Oxford before doing graduate work in mathematics, physics, and engineering, earning a PhD in Information Engineering from the University of Cambridge in 2018.

Abstract 

Uncertainty is unavoidable in the real world, and like humans, autonomous cyberphysical systems must be designed to make reliable decisions in spite of it. A central challenge is bridging the gap between model-based methods, which provide structure, interpretability, and rigorous guarantees, and model-free methods, which offer strong representation-learning capabilities and flexibility in complex systems; both properties are crucial to enabling reliable and adaptive stochastic control. This talk provides a general overview of the work that we have done and are currently doing at ACSS Lab towards this broad goal, with some emphasis on how structured representations of uncertainty can support reliable decision-making. We present several representative research thrusts including autonomous decision-making under system and environment uncertainties, data-efficient learning for adaptive autonomy under uncertainty, and large-scale robust, adaptive control for networked multiagent systems. These directions span applications in motion planning, fault/OOD tolerance, and generative AI-based robotics.

Bio 

SooJean Han is an Assistant Professor of Electrical Engineering at KAIST and the director of the Autonomous Control for Stochastic Systems (ACSS) research lab. Her research interests lie broadly at the intersection of stochastic optimal control theory and learning-enabled adaptive decision-making, with emphasis on methods that provide rigorous performance, stability, and robustness guarantees. Her lab maintains multiple interdisciplinary collaborations, including with universities in both Korea and the USA, and with agencies such as NRF, ETRI, KIAST, ADD, and AFOSR. She also serves as an AE for IEEE RA-L and IEEE IROS.  She completed her Ph.D. in Control and Dynamical Systems (CDS) at Caltech in Jan. 2023 under the guidance of Soon-Jo Chung and John C. Doyle, while supported by the NSF GRFP. She received her B.S. in Electrical Engineering and Computer Science (EECS), and Applied Mathematics at UC Berkeley in 2016.

Abstract 

We study Bayesian learning and state estimation for continuous-time linear stochastic systems with two simultaneous information constraints: the system contains an unknown parameter and the output is observed only at random sampling times. The sampling times are modeled as the jump times of a Poisson counter, and the unknown parameter is assumed to belong to a finite set of candidate values.  To account for the model uncertainty, we run a collection of Kalman-Bucy-type filters, one for each candidate parameter value, and combine their estimates using recursively updated Bayesian posterior weights. The posterior probability of each candidate model is updated from the sampled innovation likelihoods generated by the Poisson-sampled observations. We analyze the long-time behavior of this coupled learning and filtering scheme. Under an explicit posterior-consistency condition formulated in terms of Rényi-affinity bounds between the predictive innovation laws, we prove that the posterior distribution concentrates on the true parameter. We further show that the Bayesian posterior covariance functional, when evaluated along the true data-generating law, becomes asymptotically equivalent in expectation to the covariance of the correctly specified Poisson-sampled Kalman filter.

Bio 

A. Tanwani graduated from University of Illinois at Urbana-Champaign, USA in 2011 where he obtained the M.S. degree in applied mathematics, and the M.S. and Ph.D. degrees in electrical and computer engineering. He held postdoctoral positions at INRIA and Gipsa-lab in Grenoble, France, and then in the Department of Mathematics, Technical University Kaiserslautern, Germany.  Since 2015, he has been a CNRS Associate Researcher with the LAAS, Toulouse, France. His research interests include switched and hybrid systems, nonlinear control, convex analysis, systems theory, and estimation problems. Dr. Tanwani was a recipient of the Fulbright scholarship in 2006. He is currently serving as an Associate Editor for IEEE Transactions on Automatic Control since 2021, and previously served as an AE for IFAC Journal of Automatica from 2018 to 2024.

Abstract 

TBD

Bio 

TBD

Abstract 

TBD

Bio 

TBD

Abstract 

Cooperative control of a multi-channel plant is challenging when agents actuate a common system through independent local channels without knowing the plant parameters or the global input structure. This talk presents a distributed model reference adaptive control (MRAC) framework for setpoint regulation of such linear multi-channel plants. Each agent implements local reference dynamics using only its own input-channel information and neighbor communication. Through synchronization, these local dynamics collectively give rise to an emergent reference model, and the plant tracks the resulting reference trajectory via distributed MRAC. The proposed framework achieves cooperative regulation without explicit parameter identification, ensuring bounded adaptive gains while preserving a distributed information structure.

Bio 

Hyungjo Byun received the B.S. degree in Electrical and Computer Engineering from the University of Seoul, Seoul, South Korea, in 2023. He is currently a Ph.D. student in the Department of Electrical and Computer Engineering, Seoul National University, Seoul, South Korea, where he is with the Control and Dynamic Systems Laboratory (CDSL) under the supervision of Prof. Hyungbo Shim. His research interests include multi-agent systems and reinforcement learning.

Abstract 

Behavioral systems theory models dynamical systems as sets of trajectories, without committing to a particular representation. Although classical in systems theory and recently brought back into focus by data-driven control, this viewpoint remains largely unexplored as a foundation for dynamic optimization. I will argue that its representation-free nature enables scalable and modular optimization formulations, especially when model-based and data-driven system representations must coexist. The talk will develop this perspective through three case studies spanning distributed optimal control, inverse problems in signal processing, and online system identification.

Bio 

Alberto Padoan (M'10) was born in Ferrara, Italy, in 1989. He received the Laurea Magistrale degree (M.Sc. equivalent) cum laude in Automation Engineering from the University of Padua in 2013 and the Ph.D. degree in Control Theory from Imperial College London in 2018, for which his dissertation received the IET Control & Automation Doctoral Dissertation Prize (2018) and the Eryl Cadwallader Davies Prize (2020). From 2017 to 2021, he was a Research Associate in the Control Group at the University of Cambridge and a member of Sidney Sussex College. In 2021, he joined the Automatic Control Laboratory at ETH Zurich and the NCCR Automation, first as a Postdoctoral Scholar and later as a Senior Scientist. Alberto is currently an Assistant Professor in the Department of Electrical and Computer Engineering at the University of British Columbia. His research focuses on the modeling, analysis, and control of complex dynamical systems, with applications to biological and cyber-physical systems.

Abstract 

A set-valued feedback law based on nonsmooth potential-energy shaping and set-valued damping injection is presented. The controller is first defined implicitly, and the well-posedness of the closed-loop, together with the robust global finite-time stability of the desired equilibrium, are established. Subsequently, an explicit expression for the controller is derived using proximal splitting methods, leading to a digital implementation. Finally, it is shown that the proposed controller outperforms forward Euler implementations by providing higher accuracy and eliminating digital chattering.

Bio 

Félix Miranda Villatoro received the B.Eng. in Electronic Engineering from Universidad Autónoma del Estado de México (UAEMex), Mexico. He obtained the Master and Ph.D. degrees in Automatic Control from CINVESTAV-IPN, Mexico, in 2013 and 2017, respectively. From March 2017 to September 2020, he was in the Control Group of the University of Cambridge, UK. Since November 2020 he is at INRIA Grenoble-Alpes, France. His research interests include: nonsmooth dynamical systems, dissipative systems, nonlinear control, monotone operators, and non-equilibrium behaviours.

Abstract 

We address the characterization of linear systems with k-variation bounding properties, which leave the set of signals with at most k-1 sign changes invariant. Since this property is equivalent to certifying the (strict) k-sign consistency, i.e., all of the k-th order minors of the underlying system operator share the same (strict) sign, this work investigates the role of representation theory and enumerative combinatorics in making this verification tractable for Hankel matrices and operators. In the finite-dimensional matrix case, it is shown that it suffices to verify k-sign consistency on a reshaped Hankel matrix with exactly k rows, which in the infinite infinite-dimensional operator setting yields also a necessary, i.e., a complete characterization, which can be numerically verified using the so-called generalized compound systems. Comparable results were previously known only in the more restrictive setting of (strictly) k-positive Hankel structures, where all minors up to order k share the same sign. This is joint work with Prof. Tobias Damm (RPTU Kaiserslautern), as well as Zijian Liu and Tom Ashani from the Technion — Israel Institute of Technology.

Bio 

Christian Grussler received a Dipl.-Math. techn. degree (Industrial Mathematics) from TU Kaiserslautern, Germany in 2011 and an M.Sc. degree (Engineering Mathematics) from Lund University in 2012. In 2017, he received a Ph.D. degree in engineering from Lund University. From 2018–2020, he was a Research Associate at the Department of Engineering at the University of Cambridge, United Kingdom and from 2020–2021 a postdoctoral scholar at the Department of Electrical Engineering and Computer Sciences, UC Berkeley, USA. He is currently an Assistant Professor at the Stephen B. Klein Faculty for Aerospace Engineering at the Technion — Israel Institute of Technology. His research interests include generalized positive systems, model reduction, data-driven control and low-rank/sparse optimization.

Abstract 

Conic constraints to state variables naturally appear in various (mechanical, chemical, biological, etc) control systems. Usually they characterise admissible sets to some physical quantities (e.g, a safe distance, a concentration, a biomass, etc) which must be positive. This talk deals with homogeneous control systems on cones whose main feature is a scaling symmetry.  It is well known that Input-to-State Stability of any perturbed homogeneous systems can be proven considering the unperturbed system only. In this talk, the similar result is going to be reported about Input-to-State Safety (ISSf) of homogeneous systems with conic safe sets.

Bio 

Andrey Polyakov received PhD in Systems Analysis and Control from the Voronezh State University, Russia in 2005. Till 2010 he was an associate professor with this university. In 2007 and 2008, he was working for CINVESTAV in Mexico City. From 2010 up to 2013 he was a leader researcher of the Institute of the Control Sciences, Russian Academy of Sciences. In 2013 he joined Inria, Lille, France. He has co-authored more than 100 papers in peer-reviewed journals and 3 books: “Attractive Ellipsoids in Robust Control”, “Road Map for Sliding Mode Control Design”, “Generalized Homogeneity in Systems and Control”. His research interests include various aspects of nonlinear control and estimation theory, in particular, finite/fixed-time stability, generalized homogeneity, sliding mode control and Lyapunov methods for both finite-dimensional and infinite-dimensional systems.

Abstract 

A brief overview of convergence and incremental stability results is given. Next, the new existence conditions for periodic steady-state solution in time-delay convergent systems are presented. The main advantage of this result is that highly nonlinear (without meaningful linear approximation) dynamics are allowed for analysis. These conditions are developed for Persidskii and Lotka-Volterra time-delay systems.

Bio 

Denis Efimov received Ph.D. degree in Automatic Control from the Saint-Petersburg State Electrical Engineering University (Russia) in 2001, and Dr.Sc. degree in Automatic Control in 2006 from the Institute for Problems of Mechanical Engineering RAS (Saint-Petersburg, Russia). From 2006 to 2011 he was working in the L2S CNRS (Supelec, France), the Montefiore Institute (University of Liege, Belgium) and IMS CNRS lab (University of Bordeaux, France). In 2011, he joined Inria (Lille - Nord Europe center). Starting from 2018 he is the scientific head of Valse team. He is an author of more than 250 scientific articles. He is a member of several IFAC TCs and a Fellow of IEEE. He is also serving as an Associate Editor for Automatica and a Senior Editor for Systems & Control Letters.

Abstract 

MPC is a well-established control technique which relies on the iterative solution of optimal control problems (OCPs). Recently, funnel-like ideas were introduced to overcome some limitations in MPC. The latter means that "artificial" assumptions are imposed to find an initially feasible solution and to ensure recursive feasibility of MPC (i.e., solvability of the OCP at a particular time instant automatically implies solvability of the OCP at the successor time instant). It was shown that these assumptions are superfluous when "funnel-like" stage costs are introduced so that the costs grow unbounded when the tracking error approaches the funnel boundary. More precisely, in contrast to simply adding the constraints on the tracking error to the OCP with standard quadratic stage costs, funnel MPC is initially and recursively feasible, without imposing state constraints or terminal conditions and independent of the length of the prediction horizon. This is shown for a large class of nonlinear systems with relative degree one, and extensions to higher relative degree are also presented. Another recent development is the combination of funnel MPC with an additional funnel control feedback loop, which leads to a control scheme that achieves the tracking objective even in case of severe model-plant mismatches. This resolves another limitation of classical MPC: It requires a sufficiently accurate model to predict the system behavior and compute the optimal control in each step. A second extension of this approach is to improve the model by learning its parameters from data, while it is still safeguarded by the funnel controller component.

Bio 

Thomas Berger was born in Germany in 1986. He received his B.Sc. (2008), M.Sc. (2010), and Ph.D. (2013), all in Mathematics and from Technische Universität Ilmenau, Germany. From 2013 to 2018 Dr. Berger was a postdoctoral researcher at the Department of Mathematics, Universität Hamburg, Germany, and from 2019 to 2025 he was a Juniorprofessor at the Institute for Mathematics, Universität Paderborn, Germany. Since January 2026 Dr. Berger is a Heisenberg-Professor at the Institute for Mathematics, Martin-Luther-Universität Halle-Wittenberg, Germany. His research interest encompasses adaptive control, optimization-based control, differential–algebraic systems and multibody dynamics.

For his exceptional scientific achievements in the field of Applied Mathematics and Mechanics, he received the "Richard-von-Mises Prize 2021" of the International Association of Applied Mathematics and Mechanics (GAMM). Dr. Berger further received several awards for his dissertation, including the "2015 European Ph.D. Award on Control for Complex and Heterogeneous Systems" from the European Embedded Control Institute and the "Dr.-Körper-Preis 2015" from the GAMM. He serves as an Associate Editor for Mathematics of Control, Signals, and Systems, the IMA Journal of Mathematical Control and Information and the DAE Panel.

Abstract 

The confidentiality of dynamical systems is important for securing cyber-physical systems today. The breach of confidentiality is when an adversary who has compromised its sensing capabilities or its communication channel aims to infer the unmeasured states of the control system from sensor data, but has no access to actuator data. In this talk, I will present one approach towards this goal for nonlinear control systems. First, we show that the controller's state can be estimated accurately when the nonlinear closed-loop system is detectable. In the absence of detectability, controller confidentiality can still be breached with a periodic probing scheme via the sensors under a robust observability assumption, which allows for the controller's state to be estimated with arbitrary accuracy during the probing period, and with bounded error during the non-probing period. Further, stealth can be maintained by choosing an appropriate probing duration. This study shows that the controller confidentiality for nonlinear systems can be breached by balancing the estimation precision and the stealthiness of the adversary. 

Bio 

Michelle Chong is an Assistant Professor at the Department of Mechanical Engineering, Eindhoven University of Technology, the Netherlands. Michelle received the Bachelor of Engineering degree in Electrical Engineering, and the Ph.D. degree in mathematical control theory from the Department of Electrical and Electronic Engineering, the University of Melbourne, in 2008 and 2013, respectively. She has held postdoctoral positions in University of California Santa Barbara (USA), Lund University (Sweden) and KTH Royal Institute of Technology (Sweden). Michelle was the 2013 recipient of the American Australian Association's postdoctoral fellowship, and won the best paper award at the 2016 ACM/IEEE 7th International Conference on Cyber-Physical Systems (ICCPS). She is currently serving on the editorial board of the IEEE Transactions of Automatic Control, the IEEE CSS Letters, and the IFAC Journal Nonlinear Analysis: Hybrid Systems, as well as the conference editorial board of the IEEE CSS conferences. Her research interest lies in the estimation and control of hybrid dynamical systems for security, safety and privacy.  

Abstract 

Multi-agent systems with antagonistic interactions and multiple leaders generally do not achieve standard consensus unless restrictive conditions on the signed interaction graph are satisfied. Instead, such systems may exhibit richer synchronization behaviors, including bipartite consensus and containment. In this work, we propose a graph-theoretic edge-based framework for the analysis of synchronization in signed networked systems. The proposed approach is based on a signed edge agreement protocol described through signed edge-Laplacian matrices. We establish new spectral properties of signed edge-Laplacian matrices in the presence of multiple zero eigenvalues and use these results to characterize the synchronization structure of the network. In particular, we establish global exponential stability of the synchronization errors and explicitly characterize the equilibrium reached by the edge states.

Bio 

Pelin Şekercioğlu is a Postdoctoral Researcher at the Department of Decision and Control Systems at the KTH Royal Institute of Technology, Stockholm, Sweden. She obtained her Ph.D. in Automatic Control in November 2024 from Paris-Saclay University, France. Her work was carried out at ONERA - The French Aerospace Lab and L2S - Laboratory of signals and systems. She received her B.Sc. degree in Mechanical Engineering in 2019 and her M.Sc. degree in Advanced Systems and Robotics in 2021, both from the Faculty of Science and Engineering, Sorbonne University, Paris, France. In 2023, she was a visiting researcher at the University of Groningen, The Netherlands. Her research interests include multi-agent systems, signed graph theory, networked and nonlinear control with application to robotics and social networks.