23rd EANN / EAAAI 2022, 17 - 20 June 2022, Greece

Generation of Orthogonality for Feature Spaces in the Bio-inspired Neural Networks

Naohiro Ishii, Toshinori Deguchi, Masashi Kawaguchi, Hiroshi Sasaki, Tokuro Matsuo


  Neural networks are extensively developed for the deep learning and intelligent processing. To improve the performance of neural networks, the biological inspired neural networks are often studied for artificial neural networks. Models for motion processing in the biological systems have been used, which consist of the symmetric networks with quadrature functions of Gabor filters. This paper proposes a model of the bio-inspired asymmetric neural networks with nonlinear characteristics, which are the squaring and rectification functions. These functions are observed in the retinal and visual cortex networks. In this paper, the proposed asymmetric network with Gabor filters and the conventional energy model are compared from the orthogonality characteristics. To show the role of the orthogonality in the feature space, tracking characteristics to input stimulus are experimented. Then, the orthogonality basis functions create better tracking results. Thus, asymmetric structure of the network and its nonlinear characteristics are shown to be effective factors for generating orthogonality.  

*** Title, author list and abstract as seen in the Camera-Ready version of the paper that was provided to Conference Committee. Small changes that may have occurred during processing by Springer may not appear in this window.