Iris
- After doing the yeast simulations, we moved onto irises, perhaps
the most used categorisation task.
- We got good results with Pearson indicating that the solution
is somewhat general.
- We managed to throw in a third
and then a fourth subnet.
- The last subnet was used to present the category during training,
and was measured in a non-Pearson test.
- As each neuron was associated with one category, we categorised
based on the category whose neurons fired most.
- This was a bit worse (93% to 89%) than the Pearson measurement,
but reasonable.
- This removes one of the weaknesses.
