The states of a neural network convey, manipulate, and store information

If the network represents information, the network must be stable in some instance.

I. Amari-Hopfield model

复杂系统的集体行为机制：有时候比较简单

Modern Hopfield Model

drawbacks：

• 落入attractor之后难以出来
• 没有对物体之间的关系进行建模

II. Continuous Attractor Network

Neural population code：

• overlapped tuning curve covering the whole space
• largely independent response

premotor: planning

III. Continuous attractor with adaptation

控制论：负反馈

traveling wave state

oscillatory tracking

smooth tracking