Internal Seminar: Paavo Parmas, Student, Neural Computation Unit (Doya Unit)

Internal Seminar: Paavo Parmas, Student, Neural Computation Unit (Doya Unit)

Title: Model-based learning of behaviour with applications to robotics

Abstract: Recent advances in machine learning have enabled computers to autonomously learn to perform tasks by learning from data without having been explicitly told what to do or how best to achieve their goal. I will discuss how these methods can be used to allow robots to automatically learn movement control. In particular, I will present my ongoing research on a data-efficient form of learning, which consists of using data to learn a Gaussian process model of the robot's dynamics, then using this model to perform simulations to optimise the behaviour. I will briefly touch on the wider potential of applying these methods to optimise experiments in science.

 
Refreshments will be served afterwards. We hope to see you there!