Deep Machine Learning
• biologically-inspired computational intelligence approach
-massively parallel
-basic cortical circuit
-discovers structure based on regularities
-spatiotemporal dependencies
-Brain operates at 200hz ...
• goal: learning to represent the world
• hypothesis: brain represent info using repetitive hierarchical architecture
VLSI tech
• adequate tech is here
• billions of transistors on a single chip
• power requirements are low
• insensitive to yield issues
• we can build large systems today!
Ben Goertzel asks Arel what his funding requirements are ... Arel says he can build you an AGI system for 2-14 million dollars. Act now!
The hierarchical part seems very inspired/based on Numenta's work. http://numenta.com/
ReplyDeleteThey're developing a technology called HTM (hierarchical temporal memory), invented by Jeff Hawkins and developed mathematically by Dileep George.