N Value Based Logic and Neural Networks.

I had a discussion with my friend David Ha who is a Google Brain Resident at Google and he suggested that beyond Ternary logic and computing systems, N-Based logic and computing is being developed matched to the corresponding data set and application for neural networks and machine learning.  He believes strongly that neural networks and machine learning will function optimally with “non-binary codes”, by going N-valued.

Here’s some further reading:

https://arxiv.org/abs/1609.00222

Also from the ICLR, a well known conference for Machine Learning:

https://openreview.net/pdf?id=ByOK0rwlx

The great thing about N value based logic is that because of turning completeness, we can implement these systems on binary computers.  However, my feeling is if the hardware is based on ternary systems this would be optimal.

Ideally, it is possible to implement ternary computing through photonics.  The two polarizations of light representing binary and darkness for the unknown.  If we could do this, we would have neural networks computers operating at the speed of light.  I have presented this idea to my class mate and talented Prof at U of T, Joyce Poon (j o y c e . p o o n @ u t o r o n t o . c a).