A parallel island model for biogeography-based classification rule mining in julia
| Autor: | Steffen Heber, Effat Farhana, Samuel Ebert |
|---|---|
| Rok vydání: | 2018 |
| Předmět: |
0209 industrial biotechnology
Contextual image classification Computer science Biogeography Evolutionary algorithm 02 engineering and technology computer.software_genre Network topology Biogeography-based optimization Dynamic programming 020901 industrial engineering & automation Classification rule 0202 electrical engineering electronic engineering information engineering 020201 artificial intelligence & image processing Data mining Classifier (UML) computer |
| Zdroj: | GECCO (Companion) |
| Popis: | In this paper, we present a distributed island model implementation of biogeography-based optimization for classification rule mining (island BBO-RM). Island BBO-RM is an evolutionary algorithm for rule mining that uses Pittsburgh style classification rule encoding, which represents an entire ruleset (classifier) as a single chromosome. Our algorithm relies on biogeography-based optimization (BBO), an optimization technique that is inspired by species migration pattern between habitats. Biogeography-based optimization has been reported to perform well in various applications ranging from function optimization to image classification. A major limitation of evolutionary rule mining algorithms is their high computational cost and running time. To address this challenge, we have applied a distributed island model to parallelize the rule extraction phase via BBO. We have explored several different migration topologies and data windowing techniques. Our algorithm is implemented in Julia, a dynamic programming language designed for high-performance and parallel computation. Our results show that our distributed implementation is able to achieve considerable speedups when compared to a serial implementation. Without data windowing, we obtain speedups up to a factor of nine without a loss of classification accuracy. With data windowing, we obtain speedups up to a factor of 30 with a small loss of accuracy in some cases. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|