Stockholm, Sweden
Registration: https://ecc24.euca-ecc.org/conference-registration/
Time and venue: June 25, 2024 08:30 AM (Local Time) | KTH Campus Valhallavägen, Stockholm, Sweden.
Module 1 comprises three sessions, the details of which are outlined below. It is structured to provide a unifying framework, based on matrix inequality tools, for the analysis of the stability properties of NNs and for the (possibly data-based) design of control systems which include NNs as a plant under control and/or the controller.
Module 2 comprises three sessions, the details of which are outlined below. It is structured to provide a comprehensive framework for designing neural network controllers with closed-loop stability guarantees through unconstrained learning. First, we introduce a parametrization encompassing all and only those control policies that stabilize a given time-varying nonlinear system. The main insight is that we can learn over stable operators to capture all stabilizing nonlinear control policies for a wide class of nonlinear systems.
Second, we present numerically efficient and unconstrained methods to approximate NN control policies that are stabilizing by design in both centralized and distributed frameworks.
Organizers:
Giancarlo Ferrari Trecate (giancarlo.ferraritrecate@epfl.ch)
Marcello Farina (marcello.farina@polimi.it)
Alessio La Bella (alessio.labella@polimi.it)
Luca Furieri (luca.furieri@epfl.ch)
Danilo Saccani (danilo.saccani@epfl.ch)
Leonardo Massai (l.massai@epfl.ch)
The workshop is sponsored by the NCCR Automation.