Overview
With the rapid popularization of learning techniques across engineering and society, the control community has seen a growing interest in moving beyond the classical model-based paradigm. A central question is how to represent dynamical systems in a way that enables reliable decision-making and control, while retaining key control theoretic requirements such as robustness, sparsity, and uncertainty awareness.
This workshop focuses on alternative representations for control and highlights three complementary directions:
- Data-driven representations: synthesizing controllers without requiring a fully specified model while still providing theoretical guarantees, naturally promoting representations tailored to control objectives.
- Neuromorphic representations: inspired by efficient biological computation and event-based sensing, emphasizing sparse spatio-temporal representations and processing matched to event-driven signals and resource-constrained implementations.
- Extraction of essential information: representing systems through physically meaningful compressed quantities (e.g., Lagrangian/Hamiltonian structure or moments) to enable parsimonious parametrizations and resilience to outliers and distribution shifts.
These directions share a unifying theme: building representations that (i) do not require complete prior knowledge of the system, (ii) admit sparse structure in space and time, and (iii) focus on essential information for control to improve robustness and efficiency.
The workshop aims to connect these viewpoints, clarify strengths and limitations, and identify new opportunities at their intersections, bringing together experts to review recent theoretical advances and catalyze new collaborations.
Audience / prerequisites. The workshop targets both junior and senior researchers interested in theoretical advances in neuromorphic, data-driven, and moment-based methods in control. A basic background in control theory is assumed.
Expected outcomes
The workshop will aim to produce:
- A structured map of connections among data-driven, neuromorphic, and moment-based representations (commonalities, key differences, and when each is advantageous).
- A concise open-problems and research agenda compiled from talks and discussions.
- A curated reading list and slide repository (speakers’ slides, key references, and pointers to software/datasets when available), to be posted on this webpage.
- Cross-community vocabulary alignment: a short terminology note clarifying how key concepts are used across communities (e.g., representation, invariants, guarantees, event-based computation).
Tentative sechdule (August 23, 2026)
- 08:00–08:15 — Registration & coffee
- 08:15–08:30 — Welcome & opening remarks
- 08:30–10:00 — Session 1 (90 min)
- 10:00–10:15 — Coffee break
- 10:15–12:30 — Session 2 (135 min)
- 12:30–14:00 — Lunch break
- 14:00–14:15 — Coffee break
- 14:15–15:45 — Session 3 (90 min)
- 15:45–16:00 — Coffee break
- 16:00–17:30 — Session 4 (90 min)
There will be 9 presentations. Each presentation is planned as 45 minutes total (40 minutes talk + 5 minutes Q&A).
Speakers (in alphabetical order)
- Alessandro Astolfi (Imperial College London, UK)
- Anders Rantzer (Lund University, Sweden)
- Bayu Jayawardhana (University of Groningen, The Netherlands)
- Claudio De Persis (University of Groningen, The Netherlands)
- Daniel Zelazo (Technion – Israel Institute of Technology, Israel)
- Florian Dörfler (ETH Zürich, Switzerland)
- Francesco Bullo (UC Santa Barbara, USA) (tentative)
- Jin Gyu Lee (Seoul National University, South Korea)
- W.P.M.H. (Maurice) Heemels (TU Eindhoven, The Netherlands)