|SCOLARI ALBERTO||Cycle: XXXI |
Section: Computer Science and Engineering
Tutor: LANZI PIER LUCA Major Research topic
:Heterogeneous architectures for data analytics and Machine Learning applications
Advisor: SANTAMBROGIO MARCO DOMENICOAbstract:
Thanks to the unprecedented availability of data, data analytics applications are becoming central in the business and decisional processes of companies and organisations. In particular, the adoption of Machine Learning applications is currently ramping in a variety of business applications and research areas like bio-informatics and statistics. The scale of these applications, however, poses several challenges that are currently being tackled in different ways. Among the challenges, energy and power efficiency are central aspects that became fundamental in the design of these applications and of the systems executing them. Research and industry are often resorting to heterogeneous hardware solutions, which can comprise accelerators based on Field Progammable Gate Arrays or even Application-Specific Integrated Circuits. These accelerators operate in complex environments that are usually distributed and that run many applications, which should be properly interfaced with the accelerators. This heterogeneity makes design, evaluation and benchmarking very complex tasks. This works aims to explore and establish methodologies for the design of these accelerators and of the applications that use them, investigating the performance and scalability gain and how they can flexibly operate in a diverse and distributed environment. Therefore, we are investigating how to accelerate complex applications from Data Analytics and Machine Learning in order to establish the benefits heterogeneity offers to these domain.