Material Detail

This video was recorded at Machine Learning Workshop, Sheffield 2004. Gaussian process priors have recently been applied to control problems. The GPs bring advantages in their representation of model prediction uncertainty, and because the derivative of a Gaussian process is a Gaussian process, they can also incorporate derivative information, and analytically provide the uncertainty of model derivatives. This can be used to bring a natural regularization of control effort, resulting in appropriately cautious control. It can also be used to provide a 'quickened' display, which takes account of model uncertainty. I will describe our work in ambiguous or probabilistic user interfaces, and demonstrate some of the techniques we have developed for combining probabilistic models with continuous control and for providing feedback of system uncertainty to the user.

Disciplines:

• Science and Technology  / Computer Science  / Human-Computer Interaction

More...