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 1902.09240cs.LGcs.NEstat.ML
收藏

模块化作为神经网络学习中复杂性管理的一种手段

Modularity as a Means for Complexity Management in Neural Networks Learning

SIANI Institute(SIANI研究所);University of Las Palmas de Gran Canaria(拉斯帕尔马斯德大加那利大学)

David Castillo-Bolado, Cayetano Guerra-Artal, Mario Hernandez-Tejera

摘要:

训练具有大量参数或复杂架构的神经网络(NN)会产生使优化过程复杂化的不良现象。为了解决这个问题,我们提出了第一个模块化的NN设计方法,其中NN被分解为一个控制模块和几个功能模块,实现原始操作。我们通过比较单片和模块化NN在列表排序问题上的性能来说明模块化概念,并在训练速度,训练稳定性和可维护性方面显示出优势。我们还讨论了模块化NN中出现的一些问题。

英文摘要:

Training a Neural Network (NN) with lots of parameters or intricate architectures creates undesired phenomena that complicate the optimization process. To address this issue we propose a first modular approach to NN design, wherein the NN is decomposed into a control module and several functional modules, implementing primitive operations. We illustrate the modular concept by comparing performances between a monolithic and a modular NN on a list sorting problem and show the benefits in terms of training speed, training stability and maintainability. We also discuss some questions that arise in modular NNs.

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。

↑