Is Analog Computing Making a Comeback in AI Research?

[The Art of DHT]

TL;DR Summary

Anyone with experience building electronic analog computers

I am researching self-learning AI, and one of the most talked of solutions is returning to analog computing. Analog computing uses real time changes in voltage and frequency to run calculations or perform functions. Does anyone have any experience in building electronic analog computers based on the op amp?

 

[BvU]

Hello @The Art of DHT ,

$\qquad$ !​

Yes, lots of folks have lots of experience .

That may answer your question, but my guess is that it doesn't help you very much .

google is your friend. Tried some terms too ?

What is it you want to do ?

$\$

 

[Baluncore]

I am researching self-learning AI, and one of the most talked of solutions is returning to analog computing.

That is the talk, but not the practice. It is faster to emulate the AI algorithm on a digital processor. Op-amps are good at linear operations, but AI requires non-linear functions, which are more quickly and more accurately emulated by the multiply-accumulate function in a digital processor.

The first step in building an analog computer is to model the analog circuit in SPICE. Only once the algorithm and circuit has been optimised should a parallel analog solution be implemented.

 

[Rive]

AI ... one of the most talked of solutions is returning to analog computing.

The AI related analog computing (the one I know about) is not exactly the opamp kind, and honestly, I would rather take that as some kind of hybrid or unique solution than 'real' analog.

Analog computing had some strong points way back, but there was also serious drawbacks too.
By now it's simply more pain than gain.

I's just like the gold standard. Everybody knows why it was phased out, but still, some people just can't stop flirting with the idea.