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

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.