预训练神经网络中神经网络连接权重的指数离散化 To reduce random access memory (RAM) requirements and to increase speed of recognition algorithms we consider a weight discretization problem for trai