Experiments and Statistics
Experiment 1: 900 neurons, 2 spatially distinct patterns. Pattern is active for 10
cycles.
Using Pearson Product Moment Correlation.
| 0 Runs | 200 Runs | 400 Runs |
|---|
| Maximum Neurons | 10 | 30 | 31 |
|---|
| A-A Correlation | -.0074 | .7585 | .7442 |
|---|
| B-B Correlation | -.0067 | .6456 | .8077 |
|---|
| A Self Correlation | .0068 | .1649 | .2110 |
|---|
| B Self Correlation | -.0057 | .2175 | .2332 |
|---|
| A-B Correlation | -.0073 | -.0068 | -.0351 |
|---|
A-A and B-B Correlation shows reliable activation, A and B self
correlation shows persistence, and A-B correlation shows
uniqueness.
Experiment 2: 400 neurons, more acurate neural model (neurons either
inhibitory or excitatory but not both, and firing loses all activation).
Exploring different patterns. All exhaustive so no spontaneous activation
needed. Patterns varied on locality.
| Local | Half Local | Interleaved |
|---|
| Number of Runs | 300 | 350 | 1500 |
|---|
| A-A Correlation | 1 | .9286 | .9900 |
|---|
| B-B Correlation | 1 | .9492 | .9917 |
|---|
| A Self Correlation | .9970 | .8821 | .8261 |
|---|
| B Self Correlation | .9734 | .9562 | .9386 |
|---|
| A-B Correlation | -1 | -.4973 | -.9826 |
|---|
All of the patterns are learned. Distance-biasing makes
it easier to learn, but even interleaved patterns are learned.
