出版社：Academy & Industry Research Collaboration Center (AIRCC)
摘要：Clustering samples according to an effective metric and/or vector space representation is achallenging unsupervised learning task with a wide spectrum of applications. Among severalclustering algorithms, k-means and its kernelized version have still a wide audience because oftheir conceptual simplicity and efficacy. However, the systematic application of the kernelizedversion of k-means is hampered by its inherent square scaling in memory with the number ofsamples. In this contribution, we devise an approximate strategy to minimize the kernel k-meanscost function in which the trade-off between accuracy and velocity is automatically ruled by theavailable system memory. Moreover, we define an ad-hoc parallelization scheme well suited forhybridcpu-gpustate-of-the-art parallel architectures. We proved the effectiveness both of theapproximation scheme and of the parallelization method on standard UCI datasets and onmolecular dynamics (MD) data in the realm of computational chemistry. In this applicativedomain, clustering can play a key role for both quantitively estimating kinetics rates via MarkovState Models or to give qualitatively a human compatible summarization of the underlyingchemical phenomenon under study. For these reasons, we selected it as a valuable real-worldapplication scenario.
