Ai Dreams Forum
Artificial Intelligence => General AI Discussion => Topic started by: keghn on December 22, 2015, 10:45:00 pm
-
The back bone of the whole system. part one:
A deep neural network will work. One trained with with many images into it, as for an example.
A SDR matrix captures all the detected classification. If a cat is detected, in video image,
then the cat bit will be set high. If no cat is in the image, it will be set low in
the SDR. As well as features, like a
corner, or an edge.
After SDR matrix is done capturing. it is moved to another unused SDR matrix and repeats.
So the mind of the AGI will be a sequence of SDR engrams.
Then there will be a temporal NN, that looks at many at many SDRs in a row. That will make
temporal classifications and record back into a future SDR. And the temporal NN will
finding pattern loops.
Raw SDR Matrix that already exist are used to see if the correct engram course is being followed.
Completely new SDR Matrix are saved into new memory and formed into new engrams later.
SDR Matrix are made ever 1/20 of a second.
The SDR will have a linking number to the next SDR matrix. And may for a branch into
many different path, by way of motor. If it cannot by way of motors it will have a percent
chance of the direction that will be taken.
Temporal Pooling:
https://www.youtube.com/watch?v=dDtwK9JYK94 (https://www.youtube.com/watch?v=dDtwK9JYK94)
Sparse Distributed Representations, SDR matrix:
https://www.youtube.com/watch?v=iNMbsvK8Q8Y (https://www.youtube.com/watch?v=iNMbsvK8Q8Y)