In the quest of efficient hardware implementations of dynamic neural fields: An experimental study on the influence of the kernel shape

Benoît Chappet de Vangel, J. Fix
{"title":"In the quest of efficient hardware implementations of dynamic neural fields: An experimental study on the influence of the kernel shape","authors":"Benoît Chappet de Vangel, J. Fix","doi":"10.1109/IJCNN.2016.7727446","DOIUrl":null,"url":null,"abstract":"Dynamic neural field (DNF) is a popular mesoscopic model for cortical column interactions. It is widely studied analytically and successfully applied to physiological modelling, bioinspired computation and robotics. DNF behavior emerges from distributed and decentralized interactions between computing units which makes it an interesting candidate as a cellular building-block for unconventional computations. That is why we are studying the hardware implementation of DNF on digital substratum (eg. FPGA). As shown in previous papers, this implementation requires several modifications to the equations in order to obtain decent hardware surface utilisation and clock speed. Here we show that the modification of the lateral weights kernel function is possible as long as certain conditions, enumerated in Amari's seminal work are respected. Thank to metaheuristic optimisation it is possible to find the right parameters for two behavioral scenarii of bio-inspired computation interest. We show that the two most hardware-friendly kernels (difference of linear functions and piece-wise function) are as easy to tune as the traditional Mexican hat kernel. However the difference of exponential kernel is more difficult to tune.","PeriodicalId":109405,"journal":{"name":"2016 International Joint Conference on Neural Networks (IJCNN)","volume":"22 1","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"2016-07-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"1","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"2016 International Joint Conference on Neural Networks (IJCNN)","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1109/IJCNN.2016.7727446","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 1

Abstract

Dynamic neural field (DNF) is a popular mesoscopic model for cortical column interactions. It is widely studied analytically and successfully applied to physiological modelling, bioinspired computation and robotics. DNF behavior emerges from distributed and decentralized interactions between computing units which makes it an interesting candidate as a cellular building-block for unconventional computations. That is why we are studying the hardware implementation of DNF on digital substratum (eg. FPGA). As shown in previous papers, this implementation requires several modifications to the equations in order to obtain decent hardware surface utilisation and clock speed. Here we show that the modification of the lateral weights kernel function is possible as long as certain conditions, enumerated in Amari's seminal work are respected. Thank to metaheuristic optimisation it is possible to find the right parameters for two behavioral scenarii of bio-inspired computation interest. We show that the two most hardware-friendly kernels (difference of linear functions and piece-wise function) are as easy to tune as the traditional Mexican hat kernel. However the difference of exponential kernel is more difficult to tune.
寻求动态神经场的高效硬件实现:核形状影响的实验研究
动态神经场(DNF)是一种流行的皮层柱相互作用的介观模型。它被广泛地研究并成功地应用于生理建模、生物计算和机器人。DNF行为出现在计算单元之间的分布式和去中心化交互中,这使得它成为非常规计算的一个有趣的候选单元。这就是为什么我们正在研究DNF在数字基础上的硬件实现。FPGA)。如以前的论文所示,这种实现需要对方程进行几次修改,以获得体面的硬件表面利用率和时钟速度。在这里,我们表明,只要遵守Amari开创性工作中列举的某些条件,修改横向权重核函数是可能的。由于元启发式优化,有可能为生物启发计算感兴趣的两个行为场景找到正确的参数。我们展示了两个最硬件友好的内核(线性函数的差异和分段函数)与传统的墨西哥帽内核一样容易调优。然而,指数核的差异比较难以调优。
本文章由计算机程序翻译,如有差异,请以英文原文为准。
求助全文
约1分钟内获得全文 求助全文
来源期刊
自引率
0.00%
发文量
0
×
引用
GB/T 7714-2015
复制
MLA
复制
APA
复制
导出至
BibTeX EndNote RefMan NoteFirst NoteExpress
×
提示
您的信息不完整,为了账户安全,请先补充。
现在去补充
×
提示
您因"违规操作"
具体请查看互助需知
我知道了
×
提示
确定
请完成安全验证×
copy
已复制链接
快去分享给好友吧!
我知道了
右上角分享
点击右上角分享
0
联系我们:info@booksci.cn Book学术提供免费学术资源搜索服务,方便国内外学者检索中英文文献。致力于提供最便捷和优质的服务体验。 Copyright © 2023 布克学术 All rights reserved.
京ICP备2023020795号-1
ghs 京公网安备 11010802042870号
Book学术文献互助
Book学术文献互助群
群 号:604180095
Book学术官方微信