Hopfield Nets
- Hopfield Nets are a well known type of connectionist system
[Hopfield 82, Aleksander]
- Apparently they are related to spin glass problems in physics
- Consequently, there is a long history underpinning theoretical
knowledge of these Hopfield Nets
- Nodes have thresholds
- Connections are bi-directional and each has a weight
- They can be thought of as running in synchronous or asychronous
modes
- In synchronous mode
- a set of neurons is turned on
- activation is spread through the "on" connections
- if a nodes threshold is surpassed it is turned on in the next step,
and the whole process is repeated.
- Our work has focused on synchronous mode behaviour, but
Hopfield Nets can also work asynchronously
- Our CA model also works synchronously, and there are a host of
questions about switching from a step to a continuous mode.
- For example, the question of when neurons can fire again, is
not currently addressed in the CA model.
