arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2001.11853cs.LGcs.NE
收藏

Gâteaux-Hopfield神经网络方法

The Gâteaux-Hopfield Neural Network method

Universidade Federal de Minas Gerais(米纳斯吉拉斯联邦大学)

Felipe Silva Carvalho, João Pedro Braga

摘要:

本文利用线性扩展的加托导数(LEGD)建立了Hopfield神经网络(HNN)方法的一组新的微分方程。这种新方法将被称为Gâteaux-Hopfiel神经网络(GHNN)。用一阶Fredholm积分问题来验证该方法,当α > 1时,与HNN整数阶微分方程相比,该方法收敛速度快22倍.此外,通过分析不同α值的结果,观察到学习时间的限制。将指出这种新方法的鲁棒性和优点。

英文摘要:

In the present work a new set of differential equations for the Hopfield Neural Network (HNN) method were established by means of the Linear Extended Gateaux Derivative (LEGD). This new approach will be referred to as Gâteaux-Hopfiel Neural Network (GHNN). A first order Fredholm integral problem was used to test this new method and it was found to converge 22 times faster to the exact solutions for α > 1 if compared with the HNN integer order differential equations. Also a limit to the learning time is observed by analysing the results for different values of α. The robustness and advantages of this new method will be pointed out.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑