Neural Control Systems: From Mechanistic Models to Neuromorphic Feedback Controllers
Date:
Neuronal control systems: from mechanistic models to neuromorphic feedback controllers
Industrial control applications are entering a new era of development, with processing moving closer to the data and decision-making being done locally. This facilitates the seamless implementation and deployment of fast feedback loops. Feedback controllers and the algorithms they use are usually application-specific, ranging from simple, explainable, mechanistic descriptions of the control signal to complex, optimization-based or learning-based approaches. Neural networks have been shown to be a powerful and adaptive approach to controller design. Spiking neural networks, push these capabilities further, offering robustness and lightweight controllers synthesis. However, to fully realize their potential, such networks must operate natively on neuromorphic hardware, in the vicinity of the systems they sense and control. In this talk, we will explore the potential of spiking neural networks as feedback controllers and analyze the advantages in terms of energy efficiency, minimizing latency, and optimizing control performance. This is based on a number of successful case studies from real-world deployments.