Inheritances of Orthogonality in the Bio-inspired Layered Networks
Inheritances of Orthogonality in the Bio-inspired Layered Networks
复制标题
DOI：
10.1007/978-3-030-91608-4_3
复制
发表时间：
2021
期刊：
影响因子：
--
通讯作者：
Naohiro Ishii;Toshinori Deguchi;M. Kawaguchi;Hiroshi Sasaki;T. Matsuo
中科院分区：
文献类型：
--
作者：
Naohiro Ishii;Toshinori Deguchi;M. Kawaguchi;Hiroshi Sasaki;T. Matsuo
分层神经网络在机器学习、人工智能和深度学习中得到了广泛的研究。自适应机制是生物视觉网络的显著特征。本文研究了分层网络中的自适应正交性。本文提出了一种仿生非对称神经网络模型。其重要特征是视网膜和视皮层网络中的平方和矫正函数的非线性特性。结果表明，所提出的基于Gabor滤波器的非对称网络在激励条件下具有自适应正交性。在实验中，非对称网络的分类效果优于对称网络。分层非对称网络的自适应正交性继承了带Gabor滤波器的非对称网络的自适应正交性。结果表明，仿生非对称网络可以有效地生成正交性函数的基和独立子空间，这将有助于特征空间的创建和学习中的高效计算。
Layered neural networks are extensively studied for the machine learning, AI and deep learning. Adaptive mechanisms are prominent characteristics in the biological visual networks. In this paper, adaptive orthogonal properties are studied in the layered networks. This paper proposes a model of the bio-inspired asymmetric neural networks. The important features are the nonlinear characteristics as the squaring and rectification functions in the retinal and visual cortex networks. It is shown that the proposed asymmetric network with Gabor filters has adaptive orthogonality under stimulus conditions. In the experiments, the asymmetric networks are superior to the symmetric networks in the classification. The adaptive orthogonality is inherited in the layered asymmetric network from the asymmetric network with Gabor filters. Thus, it is shown that the bio-inspired asymmetric network is effective for generating the basis of orthogonality function and independent subspaces, which will be useful for the creation of features spaces and efficient computations in the learning.