New Categorisation Results
- We've been trying to develop a neuro-cognitive model of
categorisation based on Shepard, Hovland and Jenkins
paper.
- There are different types of binary category tasks with 8 members
based on 3 binary features, and people behave differently on them.
- The top is like the iris nets, but we've added pairs and triplets.
- If we put in just triplets, we get 100%, but people do better
on different category types.
- We get pretty close to that with just single inputs, and even closer
with a combination of singles and triplets.
- This is a shot of the net during a test.
- Note that neurons from both categories are firing in the output.
- Moreover, when the input stops, the output stops a few cycles
later.
