Correlation Simulation
- The simulations I'll discuss today are all based on a 20x20 net of
"neurons".
- Patterns are presented by activating neurons.
- There are two patterns, the top 11 rows and the bottom
11 rows.
- Each time 20 of the 220 neurons in the pattern are activated.
(There is no spreading activation.)
- When a neuron in the top fires, the likelihood it is connected to
another neuron in the top (or middle) is 19/219 or .0867.
![](assoc-1.gif)
- The likelihood of top to bottom is 0, and of middle to top is
.0434.
- The minimum weight is .01 so that's what it tends toward instead of
0.
- Note that the input pattern comes back later in the talk for overlapping
CAs.
- Also note that the weights just tend toward these values.