CAs vs. signal processing
- One of the weaknesses of CA models in general and CANT in particular is
that they do not currently tie up to the visual interface
- There are lots of models of how V1 neurons recognise, for instance,
line orientation.
- There has been some work using networks similar to CANT to show that
this network can be learned using Hebbian learning
- In this model neurons process signals
- CAs area a different type of model because they are looking at higher
order features (micro-features, or wicklefeatures)
- It would really strengthen the CA argument to tie the two together
to get for instance triangle and rectangle concepts.
