Slashdot Mirror
IBM Claims Big Breakthrough in Deep Learning (fortune.com)
The race to make computers smarter and more human-like continued this week with IBM claiming it has developed technology that dramatically cuts the time it takes to crunch massive amounts of data and then come up with useful insights. From a report: Deep learning, the technique used by IBM, is a subset of artificial intelligence (AI) that mimics how the human brain works. IBM's stated goal is to reduce the time it takes for deep learning systems to digest data from days to hours. The improvements could help radiologists get faster, more accurate reads of anomalies and masses on medical images, according to Hillery Hunter, an IBM Fellow and director of systems acceleration and memory at IBM Research. Until now, deep learning has largely run on single server because of the complexity of moving huge amounts of data between different computers. The problem is in keeping data synchronized between lots of different servers and processors In it announcement early Tuesday, IBM says it has come up with software that can divvy those tasks among 64 servers running up to 256 processors total, and still reap huge benefits in speed. The company is making that technology available to customers using IBM Power System servers and to other techies who want to test it.
>a subset of artificial intelligence (AI) that mimics how the human brain works.
We already have humans for that. How about doing things humans cannot do?
I should use this sig to advertise my book ISBN-13 : 978-1501515132.
Expensive, polished and flashy commercials. They should develop a server farm for rendering bullshit.
..and shot twice.
"keeping data synchronized between lots of different servers and processors"
In a parallel initiative, cryptocurrency Ethereum has begun work on its Casper platform meant to transition away from a 'proof of work' towards a 'proof of stake' incentive. Same problem set, but profits will go into your pocket, not corporations.
whatever time savings you reap will be eaten up by the time it takes to deploy IBM's crapware
We already have humans for that. How about doing things humans cannot do?
There are a lot of humans that aren't very intelligent. Furthermore just because a human can do a task doesn't mean it cannot be done better/faster/cheaper with some mechanical assistance. Humans can do remarkable things but we have our limits. This includes both physical and knowledge tasks.
How to get rich in a gold rush.
I wonder how this compares to Google's approach to speeding up ML, the Tensor Processing Unit, and whether the ideas can be combined for even faster learning.
Note to ACs: I usually delete AC replies without reading them. If you want to talk to me, log in.
"IBM (IBM, +0.44%) says it has come up with software that can divvy those tasks among 64 servers running up to 256 processors total, and still reap huge benefits in speed." Everything in this description is stuff you could do with open source code like TensorFlow 6 months to a year ago. More details are needed to call this a "breakthrough". Have they published a paper?
> ... IBM claims its system achieved 95% scaling efficiency across 256 processors ...
Sounds like bullshit to me. Or a highly optimized workload... perhaps a hand-picked problem that simultaneously maximizes per-CPU/GPU load while minimizing the need for processor-to-processor communication.
These machines:
1) Do not learn - learning requires reasoning, and these machine do not reason.
2) Do not use AI - they are not intelligent or even artificially. Again no reasoning.
These machines are get at pattern matching and rewriting a super-case statement. Finding in the pattern matching routines patterns that were not in their programs ability to "see". But as has been shown "self" driving cars, these patterns can be WRONG. Place stickers on a STOP sign, the computer can not understand it is a stop sign anymore. It is not long with in its "known" patterns of STOP sign.
So please lets starting using correct terms.. We are techies. We sould get this basic fact right. .
But beyond the end of the world jokes this stuff is still massively cool. There is a lot of really great things that can be accomplished by better AI to enrich everyone. Definitely another step towards all the neat Sci-fi things we've seen.
Uplink Hosting - Web/email at an affordable price with high performance - https://uplinkhosting.ca/link.php?id=3
IBM's "innovation" is to insert synchronizations in a render farm that enable the gathering of intermediate GPU results across a distributed batch run.
Of course, Google made the same "revolutionary discovery several years ago when Dean and crew first developed DistBelief. Later they abstracted it into TensorFlow's compute graphs. When was this, 2010?
Yeah. Another Big Blue Breakthrough.
will this AI write a memo ?
If so it doesn't need to apply at Google
Since when has a neural network been and N to M network? It doesn't have the problem they claim to solve, so I'm not clear what exactly they've invented.
Was this by anychance invented by the marketing department?
Why can't this be done in a couple shaders on a GPU?
Seriously. Does IBM actually make products anymore? "Deep Learning"?? Really? IBM, can you tell me where I can buy a Deep Learning? How about a Watson? How about a Cognitive Computing System? Can I buy a Big Data, please? From a technology standpoint, IBM has completely jumped the shark with all of this platform-y, non-productized, framework-y bullshit that requires millions in services hours to implement one-off solutions.
IBM used to make real contributions from their research division into actual software products. Postfix, anyone? RISC technology with AIX and the RS-6k was revolutionary. Their virtualization innovations became the foundation of the AS/400. But no. They jettisoned all of that.
They only do two things now: 1) Research for marketing releases to keep their stock price stable, and 2) Add cash-cow products to their portfolio through acquisition, call them "cognitive"/"big data"/"deep xxxxx", and offshore dev and tech support to a country which charges the lowest wages in the world.
They really have just went down the tubes. It is no wonder that they have declining revenues for so many quarters that I lost count.
https://xkcd.com/875/
There are several open source libraries that can distribute deep learning. There's no news here, just some drone hired by IBM to sell a fluff piece. This is why IBM is hurting these days. Open source is kicking their ass and they can't do anything about it.
Acording to the article, the main development is they use NVlink.
I.e.: they use some king of SLI bus for the interconnect between all the graphic cards in their cluster.
Instead of using OpenMPI over Infiniband like tons of scientific clusters have been doing for the past decade.
Yes, they're going to shave a tiny bit of latency (from what I gatter the interconnect is directly handled and access by the graphic cards themselves) (as opposed to Infiniband which would need to go through the bus on the motherboard).
But that's about it.
No real revolutionnary breakthrough.
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
Wow, IBM attempted to throw more hardware at the problem, and actually succeeded.
Downside: requires more hardware.
It probably does decrease average training latency at scale (not counting the shovel time invested in sunk cost), but I'd still hesitate to call this an advance.
throw more cpus at it? I'm glad they overcame any scalability problems but i can't help feeling misled by this article.
That's what they need, someone to replace their entire C team.
And sell it to GE too. Apple and Google aren't too far behind.
Reading the summary it seems that IBM have discovered distributed computing.
Releasing anything is a breakthrough at IBM. If one day the broken version of wget on AIX supports the --method option there will probably be a documentary about it on Netflix.
lucm, indeed.