PSO Efficient Implementation on GPUs Using Low Latency Memory

Abstract

This paper proposes an efficient implementation for the Particle Swarm Optimization (PSO) algorithm using the shared memory available in the Graphic Processing Units (GPU) of CUDA (Compute Unified Device Architecture) platforms. In our proposal each dimension of each particle is mapped as a thread. The threads are executed in parallel within a GPU block. Since the GPU blocks present a maximum number of allowed parallel threads, we propose to use multiple sub-swarms. Each sub-swarm is executed in a GPU block aiming at maximizing data alignments and avoiding instructions bifurcations. We also propose two communication mechanisms and two topologies in order to allow the sub-swarm to exchange information and collaborate by using the GPU global memory. The results for 8 sub-swarms, each one with 32 particles and 32 dimensions, show speedups up to 100 and 5 times when compared to the serial implementation and PSO start-of-art implementation for CUDA, respectively. Our proposal allows one to deploy PSO algorithms in continuous optimization problems, which present many input variables. This type of problem is very common in engineering.

A Fault Diagnosis Method Based on CFPN Model

“…If a transition implementation can not be disabled by an external behavior,then the transition is called uncontrollable transition.The implementation or not of an uncontrollable transition can only be determined by the structure and status of CFPN and regardless of the external environment.When given a set of constraints and there are uncontrollable transitions in FPN system,calculate control arc directly will lead to the occurrence of not allowed constraints [6] .Before design the control arc directly,the not allowed constraints need to be converted to allowed constraints.…”

Um Breve Estudo Sobre Os Algoritmos Dos Vaga-Lumes E Dos Morcegos Para Otimização De Funções De Referência

“…Os algoritmos de inteligência de enxames têm sido aplicados com sucesso em vários domínios. Podem ser citados como exemplos a) roteamento de veículos (Gambardella, Taillard & Agazzi, 1999), b) composição musical (Blackwell & Bentley, 2002), c) análise de clustering (Chen & Ye, 2004), d) aplicações de eletromagnetismo (Mikki & Kishk, 2005), e) bioinformática (Correa, Freitas & Johnson, 2006), f ) problema do caixeiro-viajante (Souza, 2013), g) treinamento de redes neurais (Zhou & Lin, 2014), h) processamento paralelo (Silva & Bastos Filho, 2015), i) otimização de funções (Paiva, Costa & Silva, 2016;Paiva, Costa & Silva, 2017), entre outros. Alguns modelos de inteligência de enxames que já foram implementados computacionalmente são: Otimização por Enxame de Partículas (Kennedy & Eberhart, 1995), Colônia Artificial de Abelhas (Karaboga, 2005), Otimização por Colônia de Formigas (Dorigo, Birattari & Stützle, 2006) e outros.…”

“…Um dos conceitos que tem sido bastante difundido está relacionado à Inteligência de Enxames. Esse conceito envolve um conjunto de meta-heurísticas cujos comportamentos emergentes podem resultar em uma capacidade de resolver problemas complexos de engenharia (Silva & Bastos Filho, 2015). Os sistemas baseados na inteligência de enxames são tipicamente constituídos por uma população de agentes simples que interagem, localmente, uns com os outros e com o seu próprio ambiente.…”

A FPN Parameter Optimization Method Based on CDEDPSO

“…Error cost function can be characterized the random uncertainty of FPN transition matrix parameters [6] . So it can take advantage of the error cost function value to reflect the strength of uncertainty parameters.The error cost function is as follows:…”