Learning for Decision and Control through the Scenario Approach

Workshop Description

This one-day workshop offers a forum for discussion on issues, challenges, and opportunities of contemporary interest regarding the wise usage of data in science and engineering. The problem is seen under the lens of the so-called scenario approach, a well-established methodology for data-driven decision-making, which has evolved over the past two decades into a comprehensive framework for using data in control design, estimation, prediction, and machine learning among others engineering applications.

In many decision-making problems, decisions are made so as to optimize performance while satisfying constraints in the presence of uncertainty. Often, the only information available about uncertainty comes from a finite set of observations, commonly referred to as scenarios. A fundamental question then arises:

Can decisions based on such limited data be trusted when facing new, unseen instances of uncertainty?

Scenario theory provides a rigorous framework to address this challenge. By offering formal risk guarantees, it turns widely used data-driven heuristics into statistically sound methods under minimal assumptions on the data-generating process.

This workshop offers an introduction to the scenario approach to learning-based decision and control for researchers new to the subject. Designed to strike a balance between foundational results and practical applications, the workshop features leading experts sharing state-of-the-art theory and algorithms related to:

a-priori guarantees and sample complexity for certified decision-making
a-posteriori risk certification, including results for non-convex problems
links between scenario theory and statistical learning theory

and with application to:

optimal stochastic constrained control in energy systems
supervised classification for medical applications
stochastic system abstraction for control design with formal guarantees

The workshop targets graduate students, control engineers, and researchers in systems & control, machine learning, and decision-making.

Key Goals

to provide an accessible entry point to state of-the-art results in the theory of the scenario approach, identifying some under-exploited results
to present practical examples illustrating the application of the scenario approach across various domains
to highlight how application needs have stimulated the growth of the theory in the past, and to identify some evolutionary forces now in action