Block-Based Neural Network Optimization with Manageable Problem Space

カテゴリ: 論文誌(論文単位)

グループ名: 【C】電子・情報・システム部門

発行日: 2020/01/01

タイトル(英語): Block-Based Neural Network Optimization with Manageable Problem Space

著者名: Kundo Lee (Mentor Graphics Japan Co., Ltd./Yokohama National University), Tomoki Hamagami (Yokohama National University)

著者名(英語): Kundo Lee (Mentor Graphics Japan Co., Ltd./Yokohama National University), Tomoki Hamagami (Yokohama National University)

キーワード: FPGA，evolvable hardware，genetic algorithm，block-based neural network

要約(英語): In this paper, a simple method based on Genetic Algorithm (GA) is proposed to evolve Block-Based Neural Network (BbNN) model. A BbNN consists of a 2-D array of memory-based modular component NNs with flexible structures and internal configuration that can be implemented in reconfigurable hardware such as a field programmable gate array (FPGA). The network structure and the weights are encoded in bit strings and globally optimized using the genetic operators. Asynchronous BbNN (ABbNN), which is a new model of BbNN, suggests high-performance BbNN by utilizing parallel computation and pipeline architecture. ABbNN's operating frequency is stable for all scales of the network, while conventional BbNN's is decreasing according to the network size. However, optimization by the genetic algorithm requires more iterations to find a solution with increasing problem space and the memory access in GA operation is one of the causes degrading the performance. ABbNN optimized with the proposed evolutionary algorithm is applied on general classifiers to verify the effectiveness with increasing problem space. The proposed method is confirmed by experimental investigations and compared with the conventional genetic algorithm.

本誌: 電気学会論文誌C（電子・情報・システム部門誌） Vol.140 No.1 （2020） 特集：電子回路関連技術

本誌掲載ページ: 68-74 p

原稿種別: 論文／英語

電子版へのリンク: https://www.jstage.jst.go.jp/article/ieejeiss/140/1/140_68/_article/-char/ja/