NextGen Synergy: Control Theory and Machine Learning

Apr 27, 2026, 8:30 AM → Apr 30, 2026, 8:00 PM Europe/Prague

Faculty of Economics and Management, Czech University of Life Sciences

Tatiana V. Guy

The inaugural conference of COST Action CA24136 NextGen Synergy: Control Theory & Machine Learning  aims to bridge the gap between Control Theory (CT) and Machine Learning (ML), two dynamic fields that are increasingly intersecting. A broad array of topics will be discussed, including but not limited to the following:

• Strengthening control-theoretical foundations
• Leveraging machine learning for control theory challenges
• Developing hybrid and data-driven models
• Translating theory into practical solutions

The conference aims to cultivate interdisciplinary collaboration among experts in mathematical analysis, numerical mathematics, control engineering, computer science, and data science. By integrating diverse knowledge, it will stimulate cross-sector dialogue and break down communication barriers.

The conference has special focus on early-career researchers (PhD students and postdoctoral fellows).
Conference webpage:  https://coml26.utia.cas.cz/

CA24136: Memorandum of Understanding

HOW TO SUBMIT TRAVEL CLAIM VIA E-COST

TRAVEL REIMBURSEMENT RULES

Mon, April 27

Registration

Opening

SESSION I

1 Learning to Control: Scalable Approximation of Value Functions for High-Dimensional Optimal Control

The curse of dimensionality in optimal control is not a wall; it is a design problem. We present a framework that addresses it through principled architecture selection, bespoke optimisation, and rigorous approximation theory. The payoff is scalable, certifiable feedback laws for mean-field and large-scale collective systems.

Speaker: Dante Kalise (Imperial College London)

2 Avoiding the Curse of Dimensionality in Structured Control Problems via Separable Neural Networks

10:40 AM Coffee Break

SESSION II

3 Control theory and adversarial machine learning

Numerous autonomous systems in areas such as social robotics, automated driving systems, or drones, are driven by decision-theoretic control models that rely on various machine learning (ML) algorithms for information processing and decision making purposes. Such ML components are susceptible of being attacked by malicious adversaries to alter the decisions made by the system in a negative manner. This is the realm of a relatively recent field of adversarial machine learning whose aim is to robustify ML algorithms against adversarial attacks. In the talk, I shall describe problems and some solutions in relation to incorporating AML algorithms into autonomous systems, with a focus on social robots as applied domain.
Based on various pieces of work with S. Liu, M. Santos, A. Nuñez, M. Chacón, T. Guy and M. Karny.

Speaker: David Rios Insua (Institute of Mathematical Sciences, ICMAT-CSIC, Spain)

4 Reduced-Order Modeling for Parameter-Dependent Optimal Control: A map-parameter-to-latent approach

5 Opponent-Aware Soft Q-Learning

12:50 PM Lunch Break

SESSION III

6 Deep Riemannian Control: Formally Verified Neural Observers and Controllers via Contraction Theory

7 Linear quadratic control of nonlinear systems with Koopman operator learning and the Nyström method

8 Deep unfolding primal/dual architectures: application to linear model predictive control

Roundtable 1: Adversarial, Risk Aware Decision Making in Multi Agent Systems.

Convener: David Rios Insua (ICMAT-CSIC)

Poster Spotlights

Poster Session A

9 An Integral Controller for Physics-Informed Learning

10 Digital Twin of a Dairy Herd: Multi-Agent Reinforcement Learning and LLM Explainability for Farm Optimisation

11 AI techniques for anomaly assessment in port structures

12 Control-Informed Machine Learning for Modeling Migration Intention Dynamics: Evidence from a Youth Emigration Survey

13 Digital Twin of the City of Zilina

14 Graph-Based Relational Learning for Cyber-Attack Detection in Water Distribution ICS: An Imbalance-Aware Study on BATADAL

15 Qualitative analysis of degenerate evolution equations arising when modelling neural, electrical and gas networks

16 Reinforcement Learning Approach for Traffic Signal Control

17 Relating Discrete-Valued Random Variables: Advanced Bayesian Structure Estimation

18 Tuning Robust Sequential Decision-Making Models in Adversarial Environments

19 On controllability of linear fractional time-varying systems and fractional evolution equations

20 Bayesian inversion methods and toolbox for their application to accidental radionuclide releases

Bayesian Inversion Methods and Toolbox for Their Application.pdf

4:50 PM Coffee Break

Excursions to Human Behaviour Research Lab & Agricultural Processing Training Centre

BEER Party

Tue, April 28

SESSION IV

21 Learning Maximal Lyapunov Functions

Lyapunov functions are essential tools in control theory for guaranteeing stability and performance. In this talk, we will introduce how to learn Lyapunov functions for finite-dimensional continuous-time dynamical systems with an application to the estimation of the maximal region of attraction.

Speaker: Matthieu Barreau (KTH Royal Institute of Technology)

22 Certifying Properties of Control Systems: Algorithms, Verification, and Theory

10:20 AM Coffee Break

SESSION V

23 Invited Talk: Machine Learning from a Control Perspective

Machine Learning has emerged as one of the most transformative forces in contemporary science and technology. In this lecture, I will discuss Machine Learning through the lens of applied mathematics, highlighting its connections with control theory, partial differential equations, and numerical analysis.

Speaker: Enrique Zuazua (Friedrich-Alexander-Universität Erlangen–Nürnberg — Alexander von Humboldt Professorship, Germany)

24 Controllability on landmark manifolds with applications to shapes and Neural ODEs

25 When beliefs about reality influence reality—May ML result in sticking in belief-distorted Nash equilibria?

12:30 PM Lunch Break

SESSION VI

26 Concentration phenomena of self-attention dynamics

27 Control of Dynamical Systems on Time Scales: Bridging Continuous and Discrete Learning Frameworks

Roundtable 2: : Machine Learning for Control. Emerging Challenges and COST Action Opportunities

Convener: Francisco Periago (TECHNICAL UNIVERSITY OF CARTAGENA)

3:40 PM Coffee Break

SESSION VII

28 Learning with Structure: How Systems and Control Theory Shapes Machine Learning for Dynamical Systems

Machine learning has emerged as a powerful tool for modeling and controlling complex dynamical systems, yet many approaches overlook structural properties with a long-standing history in systems and control theory. This plenary highlights how symmetry, invariants, Hamiltonian structure, and optimality conditions can be embedded into learning architectures for dynamical systems and optimal control. Enforcing such structure through model design, loss functions, and constraints leads to methods that are data-efficient, interpretable, and compatible with control objectives. Examples include symmetry-aware and Hamiltonian neural networks as well as optimality-informed learning for parametric control problems. Overall, the perspective bridges model-based and data-driven approaches, showing how machine learning and control theory can mutually inform and advance one another.

Speaker: Kathrin Flasskamp (Saarland University)

29 LDNets: Latent Dynamics Networks, an architecture to learn and predict time variant fields

CORE Group Meeting

Wed, April 29

SESSION VIII

30 Industrial Perspective on the Interplay between Control and Machine Learning

For decades, industrial automation has leveraged core control theory applications—such as PID control, Kalman filters, and Model Predictive Control—to significantly boost the efficiency and safety of industrial operations. On the other hand, the successful integration of Machine Learning in this domain was relatively limited until its dramatic expansion over the past decade, culminating in the recent Generative AI revolution. This presentation will explore the synergistic and occasionally antagonistic interplay between control theory and machine learning, with a specific focus on intelligent building applications. The talk will present the historical context and discuss future challenges in this evolving landscape.

Speaker: Petr Stluka (Siemens)

31 Learning from data via overparameterization

10:20 AM Coffee Break

SESSION IX

32 Online Learning in Digital Markets

Online learning explores algorithms that acquire knowledge sequentially, through repeated interactions with an unknown environment. The general goal is to understand how fast an agent can learn based on the information received from the environment. Digital markets, with their complex ecosystems of algorithmic agents, offer a rich landscape of sequential decision-making problems, characterized by diverse decision spaces, utility functions, and feedback mechanisms. This talk will demonstrate how tackling challenges within digital markets has not only advanced our understanding of machine learning capabilities but also revealed novel insights into algorithmic efficiency and decision-making under uncertainty.

Speaker: Nicolò Cesa-Bianchi (University of Milan, Italy)

33 Recursive Estimation of ARX Model with Parameters Dependent on High-Dimensional Discrete-Valued Regressors

34 A rigorous framework to certify predictions from physics-informed neural networks

Conference Photo

12:40 PM Lunch Break

SESSION X

35 A polynomial approximation scheme for nonlinear model reduction by moment matching

36 Policy gradients for deep directional reinforcement learning

37 On the Local Stabilisation of Interacting Particle Systems and its Connection to Energy Convexification

38 Mathematical Modelling and Model-Based Control Design for Energy-Optimal Motion of Industrial Robots

39 An efficient and adaptive machine learning surrogate for optimal control in enhanced oil recovery

3:40 PM Coffee Break

Roundtable 3: SWOT: Integrating Contemporary Control and Machine Learning

Poster Spotlights

5:40 PM Coffee Break

Poster Session B

40 Cone-Constrained Neural Networks for Structured Learning and Control

41 Imitation learning framework based on Fully Probabilistic Design

42 Sampling series as a unified approximation framework for control and machine Learning

43 Model Predictive Control and Machine Learning-Based Grid-Forming Inverters: The Combination of Control and Learning for Low-Inertia Power Systems

44 Learning-Phase Stability in Actor-Critic Reinforcement Learning: Analysis and Stabilisation

45 Stress Detection in the Workplace Using Digital Twin Technology and Multimodal Physiological Sensing

46 Data-Driven Identification of Port-Hamiltonian DAE Systems by Gaussian Processes

47 Decision-Focused Learning Meets Distributionally Robust Optimization: A Unified Framework for Sugar Beet Yield Maximization Under Epistemic Uncertainty

48 Multi-Teacher GNN Knowledge Transfer for Kolmogorov-Arnold Networks

49 Structural Constraints on Expressivity in Continuous-Depth Neural Networks

Poster_Pitch.pdf

50 Transcription of Czech administrative records written in Kurrent script into Latin script using machine learning, OCR, and AI

Martinek -Presentation of Transcription of Czech administrative records written in Kurrent script into Latin script using machine learning, OCR, and AI.pdf

Transcription-of-Czech-administrative-records-written-in-Kurrent-script-into-Latin-script-using-mach (4).pdf

51 Sparse Training of Neural Networks based on Multilevel Mirror Descent

52 N-to-1 Knowledge Transfer in Reinforcement Learning via Adaptive Q-function Selection

6:40 PM Free time - transfer to the conference dinner venue

Conference Dinner

Thu, April 30

SESSION XII

53 Optimization and Machine learning in Traffic and Sensor Signals

In this talk, I will present challenges that arise in an industrial setting.
1) the planning of traffic, in particular in urban areas, map matching, and optimal planing of an autonomous drive.
The major part of the talk will be on the first scenario, where we the main tension is between individual goals (often: quickest path) and global ones (reduce overall emissions in a city). This turns very quickly into a high-dimensional, multi-objective problem where machine learning can be one approach to solve it.
2) Map matching of noisy positions to a network of roads. Traditionally, this is solved by Hidden Markov Models, where the goal is to determine the optimal path on a map with positions that are noisy.
3) Deployment of AI software on hardware with restrictions (memory, cpu, scheduling involving other processes, safety, security). This will be a minor part of the talk and shall illustrate the potential that optimization and machine learning have in real-world applications. Typically packaging is done automated, for most applications that means, however, that restrictions on hardware are not considered from the beginning. The design of a feedback loop on development and efficient packaging is an example how problems may arise at places one does not think from the beginning.

Speaker: Markus Abel (Ambrosys)

54 Non-conservative Stability Analysis of Linear Systems in Feedback with ReLU Neural Networks

10:20 AM Coffee Break

SESSION XIII

55 SIMONet: a deep learning framework for learning set-valued maps. Application to Control

56 Convex synthesis of mini-batch gradient methods

Closing

11:50 AM Farewell Lunch

1:20 PM Free Time

Guided Prague Tour