When the best tool you have is a hammer make your problems in the shape of 
nails. 

Right or wrong we are going down the neural network software and hardware 
and researching spending path. GPUs give us 10-100x over CPUs, GraphCore 
and Wave Computing give another 100-200x in compute power and power 
efficiency. Maximum die size GraphCore chip in 10nm tech node with TSV 
stacked memory on top will get us a factor of 1,000,000 over CPUs. 

Folks are learning to do reasoning with nn, see reasoning with schematic 
loss function. Ben's video from Berlin 2015 about driving vision nn to how 
a more structured middle layer and Hiinton's capsules are improving vision. 
We know to use real world sequential data to do unsupervised training. For 
example books feed in a character at a time with the nn predicting the next 
letter trains for word knowledge. Books feed in a word at a time trains for 
sentence structure. The challenge seems to be to drive a rich and 
structured mid layer.   

-- 
You received this message because you are subscribed to the Google Groups 
"opencog" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
To post to this group, send email to [email protected].
Visit this group at https://groups.google.com/group/opencog.
To view this discussion on the web visit 
https://groups.google.com/d/msgid/opencog/f84b352f-160f-4355-9de9-a4f76369f784%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

Reply via email to