Banca de QUALIFICAÇÃO: ALBA SANDYRA BEZERRA LOPES

Uma banca de QUALIFICAÇÃO de DOUTORADO foi cadastrada pelo programa.
STUDENT : ALBA SANDYRA BEZERRA LOPES
DATE: 20/03/2020
TIME: 14:00
LOCAL: DIMAp
TITLE:

Using Machine Learning to Reduce Design Space Exploration Time in Heterogeneous Multicore with CGRAs


KEY WORDS:

design space exploration; heterogeneous multicore architectures; reconfigurable accelerators; machine learning; ensemble learning.


PAGES: 73
BIG AREA: Ciências Exatas e da Terra
AREA: Ciência da Computação
SUBÁREA: Sistemas de Computação
SPECIALTY: Arquitetura de Sistemas de Computação
SUMMARY:

Every year the demand of embedded applications for computational resources increases. To meet this demand, the embedded system designs have made use of the combination of diversified components, resulting in heterogeneous platforms that aims to balance the processing power with the energy consumption. However, a key question in the design of these systems is which components to combine to meet the expected performance at the cost of additional area and energy. To perform a vast design space exploration allows to estimate the cost of these platforms before the manufacturing phase. However, the number of possibilities for solutions to be evaluated grows exponentially with the increasing diversity of components that can be integrated into a heterogeneous embedded system. Evaluate the cost of one of these solutions through hardware synthesis is an extremely costly task. And even the use of high-level synthesis tools as alternative does not allow to synthesize all the solution possibilities and meet the textittime-to-market. In this work, one propose the use of prediction models based on machine learning algorithms to construct a tool for design space exploration of heterogeneous systems composed of general purpose processors and reconfigurable hardware accelerators. This tool aims to speed up the design exploration in the early stages of the design process and achieve high accuracy rates in predicting the cost of solutions. Although there are solutions in the literature that make use of the same prediction models approach, in general, these solutions address the exploration of microarchitectural parameters of only one of the components (either processors or accelerators). This work proposes the variation of the parameters of both components and also proposes the use of ensemble learning to increase the accuracy of the predictive modeling. Preliminary results show that the built prediction models are able to achieve a prediction accuracy rate of up to 98% and reduce the time for exploring the design space by 104x.


BANKING MEMBERS:
Presidente - 1882699 - MONICA MAGALHAES PEREIRA
Interno - 1694485 - MARCIO EDUARDO KREUTZ
Externo à Instituição - ANTONIO CARLOS SCHNEIDER BECK FILHO - UFRGS
Externo à Instituição - SILVIO ROBERTO FERNANDES DE ARAUJO - UFERSA
Notícia cadastrada em: 28/02/2020 16:19
SIGAA | Superintendência de Informática - | | Copyright © 2006-2020 - UFRN - sigaa02-producao.info.ufrn.br.sigaa02-producao