Slashdot Mirror
Chips That Flow With Probabilities, Not Bits
holy_calamity writes "Boston company Lyric Semiconductor has taken the wraps off a microchip designed for statistical calculations that eschews digital logic. It's still made from silicon transistors. But they are arranged gates that compute with analogue signals representing probabilities, not binary bits. That makes it easier to implement calculations of probabilities, says the company, which has a chip for correcting errors in flash memory claimed to be 30 times smaller than a digital logic-based equivalent."
It would seem that they have reinvented the analog computer, but this time entirely on a chip. And probably (hopefully) with some logic that prevents errors due to natural processes like capacitive coupling.
12.5% that understands binary 87.5 that don't...
My -1 Troll is actually a +1 funny. And my -1 flame is actually a +1 insightfull.
So basically its a computer that makes up statistical computations and corrects them to fit the models on the fly? Lazy scientists, rejoice!
The article mentions Bayesian calculations. Can these computers really speed up those calculations? Nowadays Bayesian calculations usually involve thousands of iterations of a technique called Markov Chain Monte Carlo (MCMC) unless the distributions in question are conjugate priors. The simulation then converges to the right answer.
The issue I see is that all these techniques are just math. They are either analytic (conjugate priors) or require strict error bounds in order get sensible answers (MCMC). There's no separate system of math that Bayesians use. Like many others, Bayesians just need quick reliable floating point mathematics. So anyway, I don't see how this can help Bayesian statisticians, unless it also revolutionizes engineering, physics, etc.
This is potentially a great advance. Everyone knows that analogue computing can greatly outperform digital computing (now each bit has a continuum of states so stores infinitely more data, each operation on 2 'bits'....you get the idea)....but there are many issues to resolve i.e.
1) Error correction - every 'bit' is in an erroneous state
2) Writing code for the thing - anyone got analogue algorithm design on their CVs?
How much longer before we get the "infinite improbability machine"?
"Glory is fleeting, but obscurity is forever." - Napoleon Bonaparte
as a machine learning person
This either means:
/.
You are a person who is learning from a machine or....
You are a learning machine who is now referring to itself as person! You also get excited about probabilities and you are posting on
A.I. has gone too far...
If 0.8 AND 0.6 = 0.7 (I assume you're taking the average here), then 1 AND 0 would be 0.5, when it's supposed to be 0. The only answers I would accept for 0.8 AND 0.6 are 0.6 (min) and 0.48 (multiplication). An OR gate is constructed by attaching NOT (1 - x here) gates to the inputs and output of an AND gate, yielding 0.8 or 0.92 depending on which rule you go with.
My Bistromathic drive makes that look like an electric pram
By Ben Vigoda, Co-Founder and CEO: http://phm.cba.mit.edu/theses/03.07.vigoda.pdf
Join the anonymous, help develop the network: http://www.i2p2.de
I think it is more of a probability thing then what you are thinking of. The return is the probability that the two values are the same. So 0.5 AND 0.5 would be 100% while 0.5 AND 0.6 would like 80% or something depending on the allowed error and uncertainty.
Thinking of this reminded me of BugBrain. If you want to play with Bayesian logic it has a pretty good set of examples including building a neural network to perform simple character recognition.
It's vaguely familiar, but since no two circuits are *truly* identical at the analog layer, *and* change as the temperature changes, people used digital instead where 'mostly 0' is still '0' and 'mostly 1' is still '1' regardless. Otherwise you can't mass produce them.
Of more interest is people using analog-alike bitstreams, where the average number of 1's vs 0's in a random stream is the amplitude of the analog wave. They then blend the input streams together to produce the output stream. I've mostly seen this done by Royal Holloway University to produce neural chips that *don't* need squillions of interconnections - they just blend probability streams. Looks like people are playing with optical ones now too. Why not put a story up about that instead?
It's called fuzzy logic [http://en.wikipedia.org/wiki/Fuzzy_logic].
One way to define it is NAND(x,y) = 1-MIN(x,y)
and the rest follows using usual logic rules.
I have no idea if that's what they do though.
You are a learning machine who is now referring to itself as person! You also get excited about probabilities and you are posting on /.
A.I. has gone too far...
On the plus side, it sounds like the robot revolution is going to be stymied for the same reason as my productivity. Destroy all humans! After I refresh /. one more time...
The enemies of Democracy are
If you're into the concept of fuzzy logic, then I strongly suggest reading Aldiss' Barefoot in the Head if you've not already done so.
I also recommend not reading it.
$ strings FTP.EXE | grep Copyright
@(#) Copyright (c) 1983 The Regents of the University of California.
It's got absolutely nothing to do with analog computers.
Really? Because the Fine Article from the OP, says:
Internally, Lyric's probability gates are essentially analog devices typically working with analog values called pbits that have a digital resolution of approximately 8-bits although the approach is applicable for different resolutions as well.
"[A]nalog devices working with analog values" does actually imply it is an analog computer, at least in part. Still, the overall usage sounds does novel, through the usage of Bayesian statistics "operations" logic as an alternative to the better known Boolean logic operations used in binary digital computers.
While electronic analog computers are primarily considered rare artifacts these days, analog electronics still exist, and continue to be used in various applications where an embedded computer is either overkill (no need for a re-programmable computer, application is trivial in analog), or less suitable (few very simple evaluations at very high speeds).
No, because it doesn't directly use entanglement and superposition.
Like your car uses electricity, but it's probably not an electric car.
Azural - instrumentals
"[A]nalog devices working with analog values" does actually imply it is an analog computer, at least in part. Still, the overall usage sounds does novel, through the usage of Bayesian statistics "operations" logic as an alternative to the better known Boolean logic operations used in binary digital computers.
I have to disagree with you here. An analog computer is not the same thing as analog electronics in general. As an analogy, using a few digital gates to control an alarm doesn't mean you just built a digital computer.
An analog computer is a special system that uses analog circuits to solve systems of differential equations. It is uniquely analog in the sense that it is continuous-time and has a continuously-variable output (not quantized). In the 40s and 50s, it was cheaper, more accurate (usually) and certainly required less equipment to do simulations using analog computers versus digital computers. Those days are long, long gone.
Sure, analog electronics more than "still exist". The analog IC market is growing faster than it ever has. But I would be hard pressed to call the analog subtractor in a Pipelined ADC an "analog computer", nor would I call the mixer in a mobile phone (a circuit that multiplies two analog waveforms) an "analog computer" either.