Re:What is the future of deep learning Hardware?
by petra on Sunday January 22, 2017 @07:04AM
(#13454835)
The lowest precision is using binary. Look for "binaryConnect" for research. Also the group behind pulpino did a chip implememtation of binaryConnect with great results.
Another possibility is the work of Jennifer hassler from Georgia tech on analog neural networks , including a roadmap of the possibilities.
Those are possibly the most optimal theoretically , but it's not certain that they'll work.
Another possibility is the work of Jennifer hassler from Georgia tech on analog neural networks , including a roadmap of the possibilities.
Those are possibly the most optimal theoretically , but it's not certain that they'll work.