masterThesis
Proposal of embedded standalone and distributed genetic algorithms for low-power, low-cost and small-sized memory devices
Fecha
2020-11-05Registro en:
MEDEIROS, Denis Ricardo da Silva. Proposal of embedded standalone and distributed genetic algorithms for low-power, low-cost and small-sized memory devices. 2020. 100f. Dissertação (Mestrado em Engenharia Elétrica e de Computação) - Centro de Tecnologia, Universidade Federal do Rio Grande do Norte, Natal, 2020.
Autor
Medeiros, Denis Ricardo da Silva
Resumen
This work proposes implementations of genetic algorithms targeting low-power, lowcost, and small-sized memory devices in two variants: a standalone version, to be used
in one single device, and a distributed version, to be used in multiple devices simultaneously. The motivation is to adapt and optimize this important artificial intelligence (AI)
technique so that it can be used in numerous applications where traditionally it was not
feasible to be utilized, such as in microcontrollers. An investigation about how to optimize each separated segment of the algorithm is done and extensive details about both
implementations are provided, including their source codes. Moreover, various experiments and simulations for multiple scenarios were done to validate their correct operation
using Hardware-In-Loop technique, as well as to find out limitations for the algorithm
parameters. The standalone implementation is compared to other work in the literature
and it performs faster and using fewer resources. For the distributed version, an important
analysis was done to verify that it can be faster than the standalone version but also more
power-efficient when reducing the clock speed and voltage of those devices. Finally, this
investigation could determine what is the threshold from where the distributed version,
even with a high overhead due communication between the devices, become either faster
or more power-efficient than the standalone one.