Design and HPC implementation of unsupervised Kernel methods in the context of molecular dynamics

The thesis represents an extensive research in the multidisciplinary domain formed by the cross contamination of unsupervised learning and molecular dynamics, two research elds that are coming close creating a breeding ground for valuable new concepts and methods. In this context, at rst, we describe a novel engine to perform large scale kernel k-means clustering. We introduce a two-fold approximation strategy to minimize the kernel k-means cost function in which the trade-off between accuracy and execution time is automatically ruled by the available system memory.