|
|
|
|
|
|
by Chance-Device
285 days ago
|
|
|
I think that all real progress in this space will come from materials science and optics, not computer science. The optical-electronic-optical path just doesn’t scale. You can use analog electronics all you like, but what silicon area and energy do they consume, and how does that compare with digital equivalents? How does the parts count scale with the network size and channels? When you put it all together, usually you’re way down on digital. I don’t think this is any different. Show me a way to keep the network all in optics, all the way through the network - including native positive and negative weights, weights that can actually be larger than 1, and activation functions - without any digital conversion and re-emission. |
|
|