Past event
Computer Science School Seminar Christopher Stone
Christopher Stone will present On the Power of a Single Neuron
Abstract: What can a single neuron compute? A single artificial neuron has a long-settled reputation for weakness. We show this understates the reach of one unit whose only signal-mixing operation is a higher-order tensor contraction, and which returns selected outputs and observations to its input through self-loops with delays. Trained separately on each objective, it learns Boolean functions, classifies handwritten digits, conditionally generates them as a continuous and self-repairing video, predicts a bouncing ball, and swings up and balances the chained double inverted pendulum on a cart. Control of this chaotic, underactuated system by a single unit has not, to our knowledge, been demonstrated before. The architecture is propaedeutic to neuromorphic substrates in which a high-order tensor contraction is realised in a single physical transition and recurrence is implemented by wiring alone.
Bio: Christopher Stone is an Honorary Research Fellow in the School of Computing at the University of St Andrews. Christopher's core expertise is in Artificial Intelligence, a subject he has taught in different forms since 2016. His research focuses on autonomous and self-adaptive systems, with work spanning biological simulations, physics simulation, to mathematical optimisation, with most of his publications in this area relating to Operational Research. Christopher holds an MEng in Software Engineering and received a PhD in Computer Science from Edinburgh Napier University, where he studied how to automatically synthesise problem scenarios and solvers for a broad range of combinatorial optimisation problems. He then held two research fellowships in constraint programming and continual learning at the University of St Andrews. He fosters interdisciplinary collaborations with the Physics Department on novel computational methods, such as photonic and quantum systems.