from the I'm-the-type-that-they-classify-as-quaint dept.
New Chip Can Process and Classify Nearly Two Billion Images per Second - Technology Org:
In traditional neural networks used for image recognition, the image of the target object is first formed on an image sensor, such as the digital camera in a smartphone. Then, the image sensor converts light into electrical signals, and ultimately into binary data, which can then be processed, analyzed, stored, and classified using computer chips. Speeding up these abilities is key to improving any number of applications, such as face recognition, automatically detecting text in photos, or helping self-driving cars recognize obstacles.
[...] The current speed limit of these technologies is set by the clock-based schedule of computation steps in a computer processor, where computations occur one after another on a linear schedule.
To address this limitation, [...] have removed the four main time-consuming culprits in the traditional computer chip: the conversion of optical to electrical signals, the need for converting the input data to binary format, a large memory module, and clock-based computations.
They have achieved this through direct processing of light received from the object of interest using an optical deep neural network implemented on a 9.3 square millimeter chip.
[...] "Our chip processes information through what we call 'computation-by-propagation,' meaning that, unlike clock-based systems, computations occur as light propagates through the chip," says Aflatouni. "We are also skipping the step of converting optical signals to electrical signals because our chip can read and process optical signals directly, and both of these changes make our chip a significantly faster technology."
"When current computer chips process electrical signals they often run them through a Graphics Processing Unit, or GPU, which takes up space and energy," says Ashtiani. "Our chip does not need to store the information, eliminating the need for a large memory unit."
"And, by eliminating the memory unit that stores images, we are also increasing data privacy," Aflatouni says. "With chips that read image data directly, there is no need for photo storage and thus, a data leak does not occur."
[...] "We aren't the first to come up with technology that reads optical signals directly," says Geers, "but we are the first to create the complete system within a chip that is both compatible with existing technology and scalable to work with more complex data."
[...] "To understand just how fast this chip can process information, think of a typical frame rate for movies," he continues. "A movie usually plays between 24 and 120 frames per second. This chip will be able to process nearly 2 billion frames per second! For problems that require light speed computations, we now have a solution, but many of the applications may not be fathomable right now."
Source: University of Pennsylvania
Related Stories
Photonic circuits are a very promising technology for neural networks because they make it possible to build energy-efficient computing units. For years, the Politecnico di Milano has been working on developing programmable photonic processors integrated on silicon microchips only a few mm2 in size for use in the field of data transmission and processing, and now these devices are being used to build photonic neural networks:
"An artificial neuron, like a biological neuron, must perform very simple mathematical operations, such as addition and multiplication, but in a neural network consisting of many densely interconnected neurons, the energy cost of these operations grows exponentially and quickly becomes prohibitive. Our chip incorporates a photonic accelerator that allows calculations to be carried out very quickly and efficiently, using a programmable grid of silicon interferometers. The calculation time is equal to the transit time of light in a chip a few millimeters in size, so we are talking about less than a billionth of a second (0.1 nanoseconds)," says Francesco Morichetti, Head of the Photonic Devices Lab of the Politecnico di Milano.
"The advantages of photonic neural networks have long been known, but one of the missing pieces to fully exploit their potential was network training.. It is like having a powerful calculator, but not knowing how to use it. In this study, we succeeded in implementing training strategies for photonic neurons similar to those used for conventional neural networks. The photonic 'brain' learns quickly and accurately and can achieve precision comparable to that of a conventional neural network, but faster and with considerable energy savings. These are all building blocks for artificial intelligence and quantum applications," adds Andrea Melloni, Director of Polifab the Politecnico di Milano micro and nanotechnology center.
Originally spotted on The Eponymous Pickle.
Journal Reference: Sunil Pai et al, Experimentally realized in situ backpropagation for deep learning in photonic neural networks, Science (2023). DOI: 10.1126/science.ade8450
Related: New Chip Can Process and Classify Nearly Two Billion Images Per Second
(Score: 2, Touché) by Anonymous Coward on Friday June 10 2022, @10:17PM (1 child)
If you MISrecognize 10% of those two billion images, it is 200,000,000 instances of bullshit per second, every second.
(Score: 0) by Anonymous Coward on Friday June 10 2022, @11:43PM
(Score: 0, Interesting) by Anonymous Coward on Friday June 10 2022, @10:27PM (1 child)
Truely amazing news nonetheless.
The number of applications will be limited to our imaginations.
Wonder what Black Mirror episode this will generate though.
Cause we are about to create some pretty powerful computational power into an area that will smash Moore`s Law...and that power comes with responsibility that our current human race shows no signs of taking humanitarily.
(Score: 0) by Anonymous Coward on Friday June 10 2022, @10:59PM
I'm not so sure. I think a neural net is a natural use for this because each pixel is the input to the front neurons. I can imagine the links between neurons to have their weights set by, say, the resistance of their connections and I could see this flow through "instantly." But if you need to train your net, determine your links and weights, then make your back-end chip, then it isn't a very flexible system. It will do what it was trained for very quick, but retraining it would require making another chip.
I tried to get more details, but: stupid paywall.
I did find an older article [arxiv.org] by one of the authors on arXiv, which has some details. It doesn't look as rigid as I first thought because there seems to be some variable attenuators in there, but I still need to dig through the details and see what it is about. But it is still neural nets.
(Score: 1, Interesting) by Anonymous Coward on Friday June 10 2022, @10:39PM (7 children)
Basically analog processing at its core, eh?
(Score: 2) by RS3 on Friday June 10 2022, @11:32PM
Also can compare it to asyncronous circuits / logic [wikipedia.org].
(Score: 2) by legont on Saturday June 11 2022, @12:29AM (2 children)
Yeah. The question is what exactly the chip does to the images. Separates black dots from white's? How many per image? Something more complicated? Is it even programmable? Well, more trainable)
"Wealth is the relentless enemy of understanding" - John Kenneth Galbraith.
(Score: 0) by Anonymous Coward on Saturday June 11 2022, @12:46AM
Normally, analog circuits model math functions, but if it models "neural network," I have no clue what it's "calculating." It's tough enough to figure out what digital neural-net "learns," god know what analog neural-net learns.
I'm guessing it's a black-box research - i.e, we built this analog "neural-net" chip, let's keep tweeking it till it returns the "desired result," even though we don't understand why or how.
(Score: 3, Funny) by deimtee on Saturday June 11 2022, @02:05PM
Decides whether to swipe left or swipe right.
No problem is insoluble, but at Ksp = 2.943×10−25 Mercury Sulphide comes close.
(Score: 2) by HiThere on Saturday June 11 2022, @03:51AM (1 child)
Everything is analog, if you look at it closely enough. If it segments the signals into discrete categories, then it's effectively digital, though possibly not the same digital that we're used to. (OTOH, look at the processing that happens in the optic nerve between the retina and the visual cortex. That's not the same digital that we're used to either. It's flow processing rather than store and forward.
Javascript is what you use to allow unknown third parties to run software you have no idea about on your computer.
(Score: -1, Flamebait) by Anonymous Coward on Saturday June 11 2022, @05:05AM
Shut the fuck up you clown. Go down that line of thought, everything is quantum, so far as we know, so far.
(Score: -1, Offtopic) by Anonymous Coward on Saturday June 11 2022, @06:02AM
I want the name of the ignint clown that modded (my) parent post.
(Score: 2, Offtopic) by MIRV888 on Saturday June 11 2022, @02:16AM (1 child)
Recording video at 2 billion frames per second seems like it would have a huge number of applications for research.
Also, do you want terminators? Because this is how you get terminators.
;-)
(Score: 0) by Anonymous Coward on Saturday June 11 2022, @04:59AM
> Also, do you want terminators? Because this is how you get terminators.
Good. [fandom.com]