Hardware/software co-design for Neural Network trained by improved Particle Swarm Optimization
Year: 2017
Type: dissertation
Abstract: Nowadays, using the particle swarm optimization (PSO) algorithms to train neural networks (NN) has become an attractive research.A famous method used to train the NNs is the back-propagation (BP) algorithm.Recently, previous studies have shown that the NN trained by PSO algorithms (NN-PSO) obtained higher recognition rates and lower learning errors when compared to the NN trained by the conventional BP algorithm.However, the standard PSO (SPSO) algorithm may stick to a local minimum in the training process.In this situation, the SPSO algorithm is stopped, leading to a high learning error rate or a low recognition rate of the trained NN.So far few studies have tried to overcome this problem by adding improving the standard PSO algorithms.Normally in previous studies, the improved versions of the PSO required to add more compute-intensive tasks that often increase computational burden the SPSO algorithm.Typically, previous researchers have implemented the NN trained by PSO algorithms only in hardware or only in software.Normally, the hardware implementation of the NN-PSO in these studies has been tested in a simulation using ModelSim program with the SPSO algorithm but it has not been tested in the real classification tasks.The FPGA-based NN was commonly investigated with a function-approximation, and the researchers just focused on reducing learning errors in the training phase.To deal with the issues described above, this research has three objectives.The first First of all, I would like to thank Professor Yukinobu Hoshino for his willingness to coach me, his promptness, the eye-opening discussions we had and his valuable comments on the progress of my research.In the same light, I thank my co-Supervisors, Professor Masayoshi Tachibana and Professor Yukio Mitsuyama for their excellent observations, ideas, and creativity that inspired my dissertation.I would like to thank all the professors at the Kochi University of Technology who were responsible for the Ph.D. courses that I have attended during these years.I have acquired considerable knowledge through their presentations and interactive scientific discussions.I would like to present all my gratitude to Dr. Cao Thang at the University of Tokyo for his constructive advice and interesting ideas
Author: DANG, Tuan Linh
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W2775346207
