神经网络的自旋玻璃体模型中的高阶相关

讨论神经网络的自旋玻璃体模型，首先从一个集合选出接近一个初始神经组态的图形，然后假定神经网络动力学中的神经组态，经转换神经元实现这个假设，来自干扰图形的噪声影响许多亚稳态的状况，大数目N的神经元网络存储多个图形，其大小是N个二进制数，对n＝3，网络能量最小，这个能量函数可描述三自旋相关的自旋玻璃体系，在神经元N＝20网络对不同阶进行了比较，推测三神经元相互作用在生物系统中是可实现的。

Higher-Order Correlations in Sping-Glass Models of Neural Net

We discuss Spin Glass Models of Neural net. At first,the pattern is selected out of a set,which is close to an initial neuron configuration.Then a neuron configuration is supposed in the dynamics of the neural net,and this goal is achieved by switching the neurons.The noise from the inferfering patterns is responsible for the many metastable states.The net of large numbers N of neurons stores many patterns,the size of a paterns is N binary digits.For n =3 minimizes a net energy,this energy function describes spin glass systems with correlations of up to three spins.As neurons N =20 net different orders are compared.It is speculated that 3 neuron inferactions could possibly be realized in biological systems.