The Other (Analog) Computer
이 글은 디지털 컴퓨터가 1940년대 후반 결정론적 부울 논리를 기반으로 한 단일 결정에 의해 주류가 되었으나, AI 시대에 확률적 계산을 직접 수행하는 아날로그 컴퓨터의 재등장이 주목받고 있음을 설명한다. 아날로그 컴퓨터는 물리 법칙을 이용해 낮은 에너지로 연산을 수행하지만, 노이즈와 범용성 문제로 1980년대 이후 주류에서 밀려났다. 1980년대 이후 아날로그 신경망 칩과 볼츠만 머신 같은 확률적 모델이 등장했으나, 학습 효율성 문제로 디지털 방식에 밀렸다. 최근 AI 워크로드가 확률적 계산 중심으로 전환되면서 아날로그 컴퓨팅의 가능성이 다시 부각되고 있다.
https://www.therml.ai/blog/two-tracks
#analogcomputing #digitalcomputing #ai #probabilisticcomputing #neuralnetworks
On the first weekend of May I will be back in Munich at the #VCFe. There I have an exhibition on #AnalogComputing (with a #THAT and a Leybold analog computer, using a modified #Vectrex and an HP XY #plotter as output devices) and I will give a lecture on "Unconventional Computing" (in which the analog computers also play a role).
https://vcfe.org/D/Vortraege.html#18
https://vcfe.org/D/Ausstellung.html#20
Joost Rekveld – Liberate the Machines!
https://neural.it/2026/04/joost-rekveld-liberate-the-machines/
#liberatethemachines #dialogueswithmachines #analogcomputing #cybernetics #videoart #videokunst #videoartist #videoarte #videoartinstallation #videoartwork #experimentalvideoart
here’s an interesting paper (that I’ve only skimmed and can’t really claim to have understood) that does a few interesting things.
Back when I was young and mentally agile, using parallel computers to solve Ising models was a thing, and it maybe sold a few instances of exotic machines.
This paper is part of a trend that reverses the process: here, they’re using physical instantiations of Ising models to solve parallel computations of the NP-hard variety.
That was interesting thing one.
Interesting thing two is using telecom-grade DSPs to condition the input into the system and to equalize the output, compensating for known distortions in the optical path.
Interesting thing three is what the DSPs on the output end are doing — there’s compensation for known distortions that comes out of the system, but they’re also *leaving in some of the noise*. The noise acts as “simulated”annealing (it’s more like *real* annealing).
So: they’ve gotten a lot of mileage by looking at several problems “in reverse”, which I guess is interesting thing number four.
https://arxiv.org/pdf/2509.09581
University press release with link to an article in Nature:
https://www.queensu.ca/gazette/stories/using-light-based-computing-tackle-complex-challenges
Corey Maley, philosophy professor at Purdue, talks about analog computation (45 minute video with 45 minutes of questions after):
https://coreymaley.net/about-me/
sounds like he has a book (*The analog mind *) that’s been stuck in development hell since ChatGPT sucked all the oxygen out of the area of cognition and computation.
I’m kind of sad his overview of analog computation doesn’t mention harmonic analyzers, tide prediction machines, or water models of the economy (not just a Terry Pratchett conceit!), but early in the talk he has some nice historical mechanical analog machines (multiplier, integrator, step function generator)
He’s working toward a view of computation that goes beyond Turing and Church discrete computability, but also pushing back in the analog=continuous notion
#analogComputing #cognitivescience
Running Pokémon with a Compass and Straightedge
A year ago I started building first prototypes of a device that would be capable of calculating #moon phases based on the historical Metonic cycle. The result was „Meton“, a 3D-printed lunar phase calculator. So if you need an #opendata moon machine, go ahead.
https://www.thomasweibel.ch/?post=rechner-aus-dem-drucker
Or build it yourself, from oak and steel.
#astronomy #astrophysics #mathematics #antiquity #history #3dprint #mechanics #moonphase #openglam #digitalhumanities #diy #analogcomputing
"Phương pháp máy tính tương tự mới giảm mạnh năng lượng đào tạo AI. Công nghệ mới giúp tiết kiệm hàng chục lần năng lượng cho AI. #AI #AnalogComputing #EnergyEfficiency #TríTuệNhânTạo #CôngNghệ"
sayzard
µP
Grégoire Locqueville
Neural
David Mankins
Hackaday [Unofficial]
Thomas Weibel