並列タイトル等参照テーブル分解アルゴリズムを伴うハードウェアを意識したニューラルネットワーク
一般注記Neural networks have been successfully implemented on various mobile hardware platforms. However, the high computational cost of neural networks, the large difference in computational accuracy with hardware, and the low structural similarity are often obstacles to be overcome in research.This thesis presents the knowledge and research development on the cross-application of neural networks and logic circuits, including an experimental procedure on implementation of multiple look up tables logic, an approximate decomposition method on decomposing larger size look-up tables into smaller individuals and a hardware-aware structured neural network.In summary, the cost of implementing bidirectional interaction between neural networks and logic circuits can be effectively reduced by benefiting from the logic learning capability of neural networks, the decomposition method of large-size look up tables, and the hardware-aware structure of neural networks.
コレクション(個別)国立国会図書館デジタルコレクション > デジタル化資料 > 博士論文
受理日(W3CDTF)2022-01-10T16:22:37+09:00
連携機関・データベース国立国会図書館 : 国立国会図書館デジタルコレクション