Machine Learning experts at the Katholieke Universiteit Leuven have set up a benchmarking experiment using the Microsoft Windows Azure cloud platform with the middleware of Techila Technologies. The results were compared with those obtained in a non-parallelized setup. The results show that significant analysis speed-ups can be gained when performing computational tasks in the cloud.
Machine learning (ML) techniques are becoming commonplace in business and research alike. Increasing amounts of data is being captured as a result of automatization of data collection efforts. This makes extracting insightful patterns increasingly challenging. In addition to this “data avalanche” becoming evermore overwhelming, the usage of more computationally intensive algorithms in predictive analysis tasks also gives rise to new issues and challenges, so that a Machine Learning approach typically entails a trade off between computational efficiency and predictive performance.
In recent years, however, new paradigms in analytics have been proposed geared towards solving these data and computational challenges, including cloud computing, distributed computing, and parallel computing approaches. A team headed by Professor Bart Baesens of the Department of Decision Sciences and Information Management, Katholieke Universiteit Leuven sets out to discern one of these new hypes in analytics, cloud computing, and to present a case study hereof which was performed at KU Leuven.