The possibilities of modeling neural networks in the framework of the thermodynamics of genetically disordered systems (glasses)

在遗传无序系统(玻璃)热力学框架下对神经网络进行建模的可能性

阅读:1

Abstract

Non-spin glasses possess a number of specific features which, in structural and dynamic aspects, are close to conditions necessary for neural networks to function. In a disordered network there exists a plurality of structural parameters and a number of two-level states defined by double-well potentials. Their characteristics are specified by the conditions of glass formation, i.e. by genesis. The thermodynamic description of glass as a self-organizing system (that does not require introducing an interacting potential model) leads to an unambiguous conclusion that its frequency spectrum is predetermined by the structure, which is characterized by zero-point entropy. Glass is a natural system of oscillators which form a disordered network. In this sense, glass conforms to a known model of a disordered neural network formed by interconnected oscillators. If one assumes that in living organisms the structure of a neural network (the brain) is inherited according to a genetic mechanism, the quickness of learning and recognition of patterns, the stability of associative memory and other capabilities have to be inherited genetically. The more ordered a neural network formed by distinguishable neurons, the better its capabilities.

特别声明

1、本页面内容包含部分的内容是基于公开信息的合理引用;引用内容仅为补充信息,不代表本站立场。

2、若认为本页面引用内容涉及侵权,请及时与本站联系,我们将第一时间处理。

3、其他媒体/个人如需使用本页面原创内容,需注明“来源:[生知库]”并获得授权;使用引用内容的,需自行联系原作者获得许可。

4、投稿及合作请联系:info@biocloudy.com。