E-Locus - Institutional Repository of the University of Crete

Home Collections Type of Work Post-graduate theses

Post-graduate theses

Current Record: 6518 of 6563

[Add to Basket]

Identifier

000447783

Title

The best of many worlds : efficient machine learning inference on heterogeneous hardware architectures

Alternative Title

Αποδοτική πρόβλεψη μοντέλων μηχανικής μάθησης σε ετερογενή συστήματα

Author

Τσίρμπας, Ραφαήλ Α.

Thesis advisor

Μαρκάτος, Ευάγγελος

Reviewer

Ιωαννίδης, Σωτήρης
Πρατικάκης, Πολύβιος

Abstract

Heterogeneous and asymmetric computing systems are composed by a set of different processing units, each with its own unique performance and energy characteristics. Still, the majority of current machine learning applications targets only a single device (the CPU or some accelerator), leaving the rest processing resources unused and idle. In this work, we propose an adaptive scheduling approach that supports heterogeneous and asymmetric hardware, tailored for a diversified set of machine learning models. Our scheduler can respond quickly to dynamic performance fluctuations that occur at real-time, such as data bursts, application overloads and system changes. The experimental results show that it is able to match the peak throughput of a diverse set of machine learning models, by predicting correctly the appropriate device with an accuracy of 92.5%, while consuming up to 10% less energy.

Language

English

Subject

CPU

GPU

IGPU

Issue date

2020-11-27

Collection

School/Department--School of Sciences and Engineering--Department of Computer Science--Post-graduate theses

Type of Work--Post-graduate theses

Permanent Link