MIT boffins make AI chips '1 million times faster than the synapses in the human brain'
- Reference: 1659364209
- News link: https://www.theregister.co.uk/2022/08/01/in_brief_ai/
- Source link:
A team at MIT reports that it has built AI chips that mimic synapses, but are a million times faster, and are additionally massively more energy efficient than current designs. The inorganic material is also easy to fit into current chip-building kit.
"Once you have an analog processor, you will no longer be training networks everyone else is working on. You will be training networks with unprecedented complexities that no one else can afford to, and therefore vastly outperform them all. In other words, this is not a faster car, this is a spacecraft," [1]said lead author and MIT postdoc Murat Onen.
[2]
"The speed certainly was surprising. Normally, we would not apply such extreme fields across devices, in order to not turn them into ash. But instead, protons ended up shuttling at immense speeds across the device stack, specifically a million times faster compared to what we had before. And this movement doesn't damage anything, thanks to the small size and low mass of protons. It is almost like teleporting."
[3]
[4]
Now that's some intelligent design.
Why results from machine learning models are difficult to reproduce
Princeton computer scientists Sayash Kapoor and Arvind Narayanan blame data leakage and inadequate testing methods for making machine-learning research difficult to reproduce by other scientists and say they are part of the reason results seem better than they are.
Data leakage occurs when the data used to train an algorithm can leak into its testing; when its performance is assessed the model seems better than it actually is because it has already, in effect, seen the answers to the questions. Sometimes machine learning methods seem more effective than they are because they aren't tested in more robust settings.
An AI algorithm trained to detect pneumonia in chest X-rays trained on data taken from older patients might be less accurate when it's run on images taken from younger patients, for example, Nature [5]reported . Kapoor and Narayanan believe practitioners need to clearly describe how the training and testing datasets do not overlap.
[6]
Models aren't sufficient by themselves, however, the code needs to be readily available too, they argued in a [7]paper [PDF] released on arXiv.
AI contract between Palantir and US Army Research Lab extended
The US Army Research Lab has extended its contract with Palantir to continue developing AI technologies for its combatant commands, worth $99.9 million over two years.
Both parties began working together in 2018. Palantir's software is used to build and manage data pipelines for platforms used by the Armed Services, combatant commands, and special operators. These resources, in turn, power machine learning systems deployed by various military units for combat.
"We're looking forward to fielding our newest ML, Edge, and Space technologies alongside our US military partners," Shannon Clark, senior veep of Innovation, [8]said in a statement.
"These technologies will enable operators in the field to leverage AI insights to make decisions across many fused domains. From outer space to the sea floor, and everything in-between." ®
Get our [9]Tech Resources
[1] https://news.mit.edu/2022/analog-deep-learning-ai-computing-0728
[2] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2Yuf4sYTy7asVCWkYmJ67vgAAAEw&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[3] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44Yuf4sYTy7asVCWkYmJ67vgAAAEw&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[4] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33Yuf4sYTy7asVCWkYmJ67vgAAAEw&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[5] https://www.nature.com/articles/d41586-022-02035-w?utm_source=Nature+Briefing&utm_campaign=82d0f0dfa9-briefing-dy-20220727&utm_medium=email&utm_term=0_c9dfd39373-82d0f0dfa9-42400591
[6] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44Yuf4sYTy7asVCWkYmJ67vgAAAEw&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[7] https://arxiv.org/pdf/2207.07048.pdf
[8] https://www.businesswire.com/news/home/20220728005319/en/U.S.-Army-Research-Lab-Expands-Artificial-Intelligence-and-Machine-Learning-Contract-with-Palantir-for-99.9M
[9] https://whitepapers.theregister.com/
Re: A million times faster
Fully agreed. What we lack almost entirely in our attempts at AI, is even an inkling of the organisational principles -- designed through aeons of evolutionary time, and trained through countless gazillions of lifetimes of experience -- that underpin the massively parallel computational problem-solving capabilities of natural neural systems.
Simply building bigger, faster, whizzier and bangier networks will not help until we have at least some understanding how to design and program the damn things. Convolutional Deep Networks are baby-steps along that route - not insignificant, but way, way off sufficient.
Super technology.......shame about the way the marketing comes over!
Quote: "...A team at MIT reports that it has built AI chips that mimic synapses, but are a million times faster..."
It's huge shame that this type of technology (see neural networks) is very poor at explaining just what was the logic behind a result.....
....so we know what the AI device reports as a conclusion.....but we never get to hear about the logic supporting the conclusion!!
Can real human beings get to check the logic? No!
So....unsupported conclusions....but a million times faster!!!
.....am I supposed to be impressed?
Re: Super technology.......shame about the way the marketing comes over!
That is extremely naive. Bear in mind that human beings are not necessarily that great at explaining our own "logic".
For instance, can you explain to me precisely the logic you used to pick your sister's face out in a jostling crowd?
Another example: I am a mathematician. When I arrive at a new approach to attacking a problem, chances are I couldn't for the life of me explain how I got there - it seemed to crystallise out of a mess of half-formed, nebulous and abstract ideas floating around in my head (apropos of nothing, this usually happens in the shower, after a good night's sleep).
When a musician composes (or even just plays) a piece of music, do you imagine they can explain to you the logic behind accomplishing those things?
Do you think it is even possible to trace the dynamical logic behind the extraordinary feats of aerial manoeuvring performed by a bee, a housefly or a bat?
If you want a system which simply chugs through clearly differentiated logical pathways, you are talking about "expert systems". You can certainly explain how they arrive at a result - problem is expert systems turned out to be pretty rubbish at dealing with real-life problems (look up "GOFAI" - Good Old-Fashioned AI). That project died in the 80s, floundering in an ocean of combinatorial explosions. It is not how nature solves complex problems.
Future AI is not going to look like an expert system, and it is not going to tell you (or not very well, at any rate) how it does what it does. Get used to that.
A million times faster
From the press release " arrays of programmable resistors in complex layers "
At last an approximation to the way the brain is physically constructed (subject to how interconnections are made and pruned, of course). However even given that, speed is not by any means the most important factor. Human synapses operate in the millisecond range, but that hasn't stopped the human brain at best coming up with incredible ideas and discoveries (including this one). We ought to consider (as a minimum) [a] the evolution of the brain in the context of human cultures and [b] the truly vast amount of incredibly diverse information that each individual brain is exposed to from (possibly even before) birth to the point where the human becomes quasi-autonomous. So there's much more to human intelligence that the wiring and speed of signalling. Nevertheless this is a potentially important contribution towards competent simulation of brain activity.