Final Report
Dates and location
8-10 June 2026, Paris, France
Chairperson
Onofrio Semeraro
Co-chairperson
Stefano Discetti, Lionel Mathelin
Conference fees
- Early-bird registration: 350.00 €
- Late registration: 400.00 €
What other funding was obtained?
Campagne Dialog-2026
CNRS-Sciences informatique
What were the participants offered?
Provision of a lecture theatre, book of abstracts, breakout spaces, and catering for the five coffee breaks. Lunches were provided at the CNAM restaurant on all three days of the event.
The afternoon of the second day was dedicated to social activities. Participants were offered a visit to the CNAM Museum, followed by the colloquium dinner at a restaurant in the vicinity of the main venue.
A special issue of the European Journal of Mechanics B/Fluids is planned, and selected participants may be invited to submit manuscripts relevant to the aims of the colloquium
for consideration.
Applicants (members)
- Guy Y. Cornejo Maceda
- Francesco Fossella
- Colin Leclercq
- Luigi Marra
- Defne Ege Ozan
Applicants (non members)
- Alessandro Bongarzone
- Marco Castelletti
- Vaibhav Chaturvedi
- Valerio Di Domenico
- Stefano Discetti
- Olga Fink
- Alessandro Franchini
- Philippe Gilotte
- George Haller
- J. Simon Kern
- James King
- Lukasz Klotz
- Aymane Lahgazi
- Eloi Lebrec
- Yannick Lecomte
- Felice Manganelli
- Andrés Marcos
- Lionel Mathelin
- Miguel Alfonso Mendez
- Ulrich Junior Meukiedje Kamgang
- Babak Mohammadikalakoo
- Iraj Mortazavi
- Gerardo Paolillo
- Luc Pastur
- Jingran Qiu
- George Rigas
- Alicia Rodríguez Asensio
- Carlos Sanmiguel Vila
- Kushagra Saurabh
- Taraneh Sayadi
- Lorenzo Schena
- Onofrio Semeraro
- Denis Sipp
- Guillermo Suarez
- Iacopo Tirelli
- Matteo Tomasetto
- Niccolo' Tonioni
- Junjie Zhang
- Zisong Zhou
Scientific report
The colloquium was motivated by recent advances in data-driven methods enabled by progress in experimental hardware, computational power, sensing technologies, and actuation strategies. Reduced-order modelling, enriched by data-driven closures and neural network approaches, has significantly improved the ability to obtain accurate and computationally efficient predictions while retaining physical interpretability. At the same time, reinforcement learning has emerged as a promising framework for the control of nonlinear dynamical systems by learning control policies directly from interactions with complex systems. The event brought together researchers working at the interface of these complementary approaches, providing a forum to discuss recent theoretical developments, computational methods, and applications. The scientific exchanges highlightedchallenges, including model robustness, generalization, interpretability, and the integration of data-driven methods with established control theory.
The programme brought together 45 participants from 10 countries and was organised around three main scientific themes.
Data-driven modelling and reduced-order models – Reduced-order modelling has become an essential tool for the analysis and simulation of complex dynamical systems.
The programme covered recent developments in Koopman-based methods, the renewed interest in the Mori–Zwanzig formalism, and machine learning approaches for closure modelling and state reconstruction including generative AI methods as well as theoretic tools from control like Wiener filtering.
Model-based control – The second theme focused on model-based control strategies for nonlinear dynamical systems. Contributions addressed optimal control, model predictive control, observer design, and estimation techniques based on reduced-order and physics-based models. Particular attention was devoted to the interaction between modelling and control, illustrating how recent advances in reduced-order modelling and data assimilation have strengthened the reliability and computational efficiency of established control methodologies.
Model-free control and reinforcement learning – The third theme explored model-free approaches, with particular emphasis on reinforcement learning and data-driven control. Alongside numerical and experimental studies, the programme also included presentations addressing theoretical aspects of reinforcement learning and its relationship with classical optimal control and dynamic programming, with a focus on reinforcement twinning, combinations of model predictive control and reinforcement learning, data-driven optimal control methods, providing a perspective on the opportunities and current limitations of these methods.
The scientific programme comprised 34 contributed presentations, each allocated 20 minutes, together with four keynote lectures. The keynote speakers were selected to represent the three principal themes of the colloquium while illustrating their impact on practical scientific and engineering applications.
The colloquium opened with a keynote lecture by Prof. George Haller entitled Data-driven Modeling and Control via Spectral Submanifolds. The lecture presented recent developments in the data-driven identification of reduced dynamics on spectral submanifolds, providing a mathematically rigorous framework for constructing predictive reduced-order models without explicit knowledge of the governing equations. The presentation also discussed the relationship between this methodology and reinforcement learning, highlighting complementary perspectives for modelling and control. Later on the first day, Prof. Olga Fink presented Dynami-CAL GraphNet, a momentum-conserving physics-informed graph neural network framework for reduced-order modelling of dynamical systems. The methodology was illustrated through applications including human motion prediction and molecular dynamics simulations, demonstrating its versatility across different physical systems and its potential for real-time, scalable, and generalizable modelling.
The second day opened with a keynote lecture by Prof. Denis Sipp entitled Koopman-based Model-Order Reduction: Coherent Structures, Frequencies and Damping Rates in Stochastic Cavity Flow. The presentation brought together several important developments in modern fluid mechanics, including the Koopman framework, the Mori–Zwanzig formalism, Ruelle–Pollicott resonances, and spectral proper orthogonal decomposition. The methodology was illustrated using a stochastic cavity flow and highlighted recent advances in the identification and interpretation of reduced dynamical models.
The concluding keynote lecture was delivered by Prof. Elie Hachem and focused on the integration of computational fluid dynamics, advanced numerical methods, and reinforcement learning for fluid and energy-related applications. The presentation demonstrated how data-driven optimisation strategies could be coupled with high-fidelity numerical simulations through examples including industrial furnace optimisation and the design of photovoltaic installations under challenging operating conditions, before outlining future perspectives for data-driven control of complex engineering systems.
All scientific sessions were held in the Amphi Gaston Planté at the CNAM in Paris. The adjacent spaces served as breakout areas and hosted the coffee breaks, while lunches were provided at the CNAM restaurant. The afternoon of the second day was devoted to social activities, including a guided visit to the CNAM Museum, followed by the conference dinner at a nearby restaurant. These activities provided additional opportunities for scientific exchange and informal discussions among participants.
Taken together, the keynote lectures provided an overview of recent developments across the three scientific themes of the colloquium. Combined with the contributed presentations, they stimulated sustained scientific discussions throughout the meeting and contributed to a programme that was scientifically rigorous and highly interactive. The balance between theoretical contributions, methodological developments, and application-oriented presentations fostered productive discussions throughout the meeting and encouraged interactions across disciplines.
Beyond the scientific programme itself, the colloquium created opportunities for establishing new collaborations and strengthening existing ones between participants from different institutions and countries. The organisers expect that these exchanges will continue through future collaborations and joint research activities.To further disseminate the scientific contributions presented during the colloquium, a special issue of the European Journal of Mechanics B/Fluids is planned, and selected participants are expected to be invited to submit manuscripts aligned with the scientific themes of the colloquium.
Number of participants from each country
| Country | Participants |
|---|---|
| France | 18 |
| Spain | 7 |
| Italy | 4 |
| Belgium | 4 |
| Switzerland | 3 |
| United Kingdom | 3 |
| Germany | 2 |
| Poland | 1 |
| Netherlands | 1 |
| Sweden | 1 |
| Total | 44 |