The idea of considering analog models of computation (by opposition to today’s digital models) is not new, and actually the first ever built computer were analog. Some well-known models of analog computers have been proposed, such as the GPAC (General Purpose Analog Computer) of Claude Shannon in 1941. However, these models have been mostly forgotten as they were 1) thought to be less powerful than a modern computer 2) not as good as modern computers for doing precise computations.
But it turns out that 1) is wrong, and this was proved only a decade ago, and analog computers could even solve some problems faster and 2) that many modern applications of computers are precisely contexts, such as machine learning or deep learning, where precision is not so important. What matters most is speed and energy consumptions, which are precisely the strength of analog computers.
Actually, analog computing comes historically also from the idea of computing by analogy. We will review various recent rebirth of analog computing and analog models, including recent startups proposing very fast computing for deep learning, or (re)birth of models based on analog computing, such as neural ordinary differential equations, or explanations of performances of some digital models through the eyes of analog computing.