Also, if you mean how would Convolution and Pooling and RELU be done, that can
be done in my design by time delay of individual nodes firing - if "the cat ate
food" is heard it may activate features "the cat" and "ate food" which may
activate the node "the cat I saw ate food". Little parts get matched, sometimes
not fully as shown because of time delay/ order is different, which ni turn
these partially activated nodes activate yet higher nodes partially too!
Basically it's ok if not exact match, order of parts of parts of parts is
resultingly similar. For Pooling, we can reuse an already existing mechanism -
when prediction candidates are activated in my hierarchy design, only 1 usually
top prediction is heard/spoken, meaning even though others do have energy, they
get a lot less weight. And RELU is exactly that.
------------------------------------------
Artificial General Intelligence List: AGI
Permalink:
https://agi.topicbox.com/groups/agi/Tf06e133ecd7df7c9-Md876f677c00294a97af65e7d
Delivery options: https://agi.topicbox.com/groups/agi/subscription
