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 2011.11713cs.LGcs.AImath.STstat.TH
收藏

偶激活函数在神经网络中的应用

A Use of Even Activation Functions in Neural Networks

University of Idaho(爱达荷大学)

Fuchang Gao, Boyu Zhang

摘要:

尽管人们对将深度学习技术应用于科学发现有着广泛的兴趣,但由于可能的函数的广阔前景和深度神经网络的“黑匣子”性质,学习准确描述科学数据的可解释公式非常具有挑战性。成功的关键是有效地将关于数据底层结构的现有知识或假设整合到深度学习模型的架构中,以指导机器学习。目前,这种集成通常通过定制损失函数来完成。在这里,我们提出了一种替代方法来整合现有的知识或假设的数据结构,通过构建自定义的激活函数,反映这种结构。具体来说,我们研究了一种常见的情况下,从数据中学习的多元目标函数$f$是部分可交换的,\emph{i.e.} $f(u,v,w)=f(v,u,w)$为$u,v\in \mathbb{R}^d$。例如,对于在左右翻转下不变的图像的分类,满足这些条件。通过理论证明和实验验证,我们表明,使用一个甚至激活函数在一个完全连接的层提高神经网络的性能。在我们的实验9维回归问题中,用指定的“海鸥”激活函数$\log(1+x^2)$替换其中一个非对称激活函数会导致网络性能的大幅改善。令人惊讶的是,即使是激活函数也很少在神经网络中使用。我们的研究结果表明,定制的激活函数在神经网络中有很大的潜力。

英文摘要:

Despite broad interest in applying deep learning techniques to scientific discovery, learning interpretable formulas that accurately describe scientific data is very challenging because of the vast landscape of possible functions and the "black box" nature of deep neural networks. The key to success is to effectively integrate existing knowledge or hypotheses about the underlying structure of the data into the architecture of deep learning models to guide machine learning. Currently, such integration is commonly done through customization of the loss functions. Here we propose an alternative approach to integrate existing knowledge or hypotheses of data structure by constructing custom activation functions that reflect this structure. Specifically, we study a common case when the multivariate target function $f$ to be learned from the data is partially exchangeable, \emph{i.e.} $f(u,v,w)=f(v,u,w)$ for $u,v\in \mathbb{R}^d$. For instance, these conditions are satisfied for the classification of images that is invariant under left-right flipping. Through theoretical proof and experimental verification, we show that using an even activation function in one of the fully connected layers improves neural network performance. In our experimental 9-dimensional regression problems, replacing one of the non-symmetric activation functions with the designated "Seagull" activation function $\log(1+x^2)$ results in substantial improvement in network performance. Surprisingly, even activation functions are seldom used in neural networks. Our results suggest that customized activation functions have great potential in neural networks.

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。

↑