Johan Suykens, UK Leuven

05-04-2022

Neural networks, kernel machines and duality principles

Abstract:

Neural networks and deep learning have attracted much attention, serving as powerful parametrizations for nonlinear functions in many different applications. On the other hand solid foundations have been established with support vector machines and kernel-based approaches in statistical learning theory and optimization. In this talk we discuss function estimation and model representations with respect to duality principles. It enables to obtain new synergies between neural networks, deep learning and kernel machines. It will be explained both for supervised and unsupervised learning problems, and further extended to generative models, multi-view learning, and deep kernel machines.