PhD position in Systems and Control Theory for Energy-based Learning

About this position

Are you excited about developing new mathematical foundations for energy-efficient computing? Do you want to contribute to cutting-edge research at the intersection of systems and control theory, optimization, circuit theory, and neuromorphic computing?

The University of Groningen is seeking a highly motivated PhD candidate to work on a fundamental research project on systems and control theory for learning in neuromorphic circuits. Neuromorphic computing is an analog, brain-inspired computing paradigm with the potential to drastically reduce energy consumption while enabling faster inference than conventional digital architectures. A major challenge, however, is the development and analysis of dedicated algorithms for training analog circuits directly from data.

In this PhD project, you will develop a novel system-theoretic framework for learning in analog circuits and dissipative networks. We will view learning as a feedback interconnection of continuous-time (circuit) dynamics and an optimization algorithm. The key idea is to develop algorithms that minimise cost functions inspired by notions of energy, leading to highly efficient, local learning rules.

What are you going to do?

As a PhD candidate, you will develop mathematical theory for learning in nonlinear and dynamic circuits. Building on preliminary results for resistive circuits, you will study circuits containing memristive and capacitive elements, as well as more general dissipative networks. The project combines systems and control theory, circuit theory, optimization, and machine learning, with the ultimate goal of advancing the mathematical foundations of physics-based learning.

Your responsibilities include:

• Developing a system-theoretic framework that models learning as the feedback interconnection between continuous-time circuit dynamics and optimization algorithms.
• Designing novel energy-based learning algorithms for training analog circuits directly from input-output data.
• Developing fully decentralised learning rules that rely on local circuit information and are suitable for large-scale systems.
• Establishing rigorous theoretical guarantees for convergence and scalability of the proposed learning algorithms.
• Extending the theory from analog circuits to more general dissipative networks.
• Testing and validating the developed methods.
• Publishing research findings in leading international journals and conferences and presenting your work at scientific meetings.
• Contributing to teaching activities and supervising Bachelor's and Master's students where appropriate.