The objective with this project is to provide design tools and algorithms for model management in robust, adaptive and autonomous engineering systems. The increasing demands on reliable models for systems of ever greater complexity have pointed to several insufficiencies in today’s techniques for data-driven model construction. The proposal addresses key areas where new ideas are required. Modeling is a central issue in many scientific fields. System Identification is the term used in the Automatic Control Community for the area of building mathematical models of dynamical systems from observed input and output signals. Several other research communities work with the same problem under different names, such as (data-driven) learning. We have identified five specific themes where progress is both acutely needed and feasible:
Encounters with Convex Programming Techniques: How to capitalize on the remarkable recent progress in convex and semidefinite programming to obtain efficient, robust and reliable algorithmic solutions.
Fundamental Limitations: To develop and elucidate what are the limits of model accuracy, regardless of the modeling method. This can be seen as a theory rooted in the Cramer-Rao inequality in the spirit of invariance results and lower bounds characterizing, e.g., Information Theory.
Experiment Design and Reinforcement Techniques: Study how well tailored and “cheap” experiments can extract essential information about applications-relevant system properties. Also study how such methods may relate to general reinforcement techniques.
Potentials of Non-parametric Models: How to incorporate and adjust techniques from adjacent research communities, e.g. concerning manifold learning and Gaussian process regression in machine learning.
Managing Structural Constraints: To develop structure preserving identification methods for networked and decentralized systems.
This is a collaborative project involving the Division of Decision and Control Systems, School of Electrical Engineering and Computer Science, KTH and the Division of Automatic Control, Department of Electrical Engineering at Linköping University.
Division of Decision and Control Systems KTH
Division of Automatic Control, Linköping University
Project funding and duration
This is project is funded by an ERC Advanced Grant with duration 2011-2015.