智能控制——神经网络控制4.pdf

本节内容 本节内容 3.4 Hopfield 网络 3.4.1 离散Hopfield 网络 1 网络结构和工作方式 2 稳定性和吸引子 3 连接权的设计 4 举例 5 matlab 的Hopfield 函数 3.4.2 连续Hopfield 网络 1 网络结构和工作方式 2 优化计算举例 (TSP 问题) 联想记忆 联想记忆 Store a set of patterns in such a way that when presented with a new pattern, the network responds by producing whichever one of the stored patterns most closely resembles. Partial content can act as the address to the full content. on en -a ress memor es n -a ress memor es n physical systems physical systems Physical systems can also act as a content-addressable memory (CAM). Consider the time evolution of the system in its state space. Some systems can be configured to have locally stable points. If the system is released from an initial state then it will roll down to the closes stable state. Defining CAM Defining CAM CAM can be defined as a system whose stable points can be set as a set of pre-defined states. The stored patterns divide the state space into locally stable points, called “basins of attraction” in dynamical Hopfield model Hopfield model Neural networks and physical systems with emergent collective computational J.J. Hopfield properties , Hopfield and Tank, PNAS, 1982. Hopfield网络结构 Hopfie