10-15 yrs ago, appears ~everybody talking cortical implementation of belief prop, ~nobody talking cortical implementation of back prop: https://scholar.google.com/scholar?q=related:CP4ffwsbGlcJ:scholar.google.com/&scioq=Cortical+circuitry+implementing+graphical+models&hl=en&as_sdt=0,5 … (not saying this is bad -- both very inspirational algorithms -- just interesting)
-
-
I don't really have the experience or seniority to say which way is objectively better. But I have worked at an embedded hardware company, so I can say this: mass-manufacturing a fully programmable approximate inference chip and getting it in affordable boxes is the hard part.
-
Even putting together established IP blocks on a novel board and writing the minimal firmware to drive the result takes half a decade of work and many tens of millions of dollars. And it will often fail, because of mismanagement or the power requirements of the chip.
- 6 more replies
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.