Slashdot Mirror
Is Microsoft Mainstreaming Machine Learning? (networkworld.com)
Tuesday Microsoft updated their open source Microsoft Cognitive Toolkit (CNTK), adding support for both C++ and Python. "This announcement is more than a point release..." argues Network World. "It's the recognition of AI and machine learning as the next big platform after mobile."
This announcement represents a shift in Microsoft's customer focus from research to implementation... The toolkit is a supervised machine learning system in the same category of other open-source projects such as Tensorflow, Caffe and Torch. Microsoft is one of the leading investors in and contributors to the open machine learning software and research community. A glance at the Neural Information Processing Systems conference reveals that there are just four major technology companies committed to moving the field of neural networks forward: Microsoft, Google, Facebook and IBM.
A Microsoft engineer described CNTK as "democratizing AI," according to Microsoft's announcement, which also notes that their toolkit "has been optimized to best take advantage of the NVIDIA hardware and Azure networking capabilities that are part of the Azure offering."
A Microsoft engineer described CNTK as "democratizing AI," according to Microsoft's announcement, which also notes that their toolkit "has been optimized to best take advantage of the NVIDIA hardware and Azure networking capabilities that are part of the Azure offering."
They get cheaper & more shady by the day. This is just a way to get to their stupid AI spyware trained, built and wedged into every MS orifice.
Bet it has onerous licensing terms too.
Seriously, mainstreaming? That's a word now?
Can we please get some technical words on a technical site and not a bunch of hipster wanna MBA jargon?
TensorFlow is what everyone is using because it works well and it has a nice license to go with it. Besides, willingly becoming reliant on anything Microsoft makes is a devil's bargain.
Anons need not reply. Questions end with a question mark.
Deep learning does not constitute all of, or even a significant chunk of, machine learning. It is merely the latest fad, much like personalized medicine in medical informatics or nano-machines in biochemistry.
(I'd even go as far as to say that it's one of the worst parts of the field since neural network models are prohibitively hard to interpret and draw conclusions from)
that this would not be unprecedented. After all, Microsoft was an absolutely essential player in "mainstreaming" personal computing. For better or worse, before PC "clones" arrived on the scene running MS-DOS and then Windows, computing was *very* expensive and not mainstream at all. This is in no way to defend other business practices MS has had over the years, but a careful look at the record should show that MS was key to the "PC revolution".
I'll show myself the door now, thanks.
Forget Google and Facebook, a multitude of smaller companies have been there before Microsoft arrived today and "gifted" the world with their ML toolkit.
- files
- deleted files
- user activity
- camera / microphone input
has been sent to Michrosopht CLOUD Services, where it can be safely reviewed, shared with 3rd parties, shared with 4th parties, breached by nebulous state actors, inadvertently corrupted and/or irretrievably lost as per the 270-page EULA designed to be so obtuse you ignore it.
Thank you, for chosing Microsoft.
if you want to make racist twitterbots. ;)
Anons need not reply. Questions end with a question mark.
No, and stop posting ads for Microsoft as if they were news articles.
It's like thinking that having a compiler means that I can program.
Has anyone else heard this song and dance before? Machine learning has been around since at least the late 1950s and the days of the perceptron systems. By the end of the 1960s it was clear to just about everyone involved that perceptrons and machine learning weren't going to deliver on the lofty hopes and grand promises of the early AI enthusiasts. Over the following decades machine learning came and went several times more. There were the "expert systems" of the 1980s, the neural networks of the 1990s and now comes "big data" and "the cloud". Are we getting the pattern yet? About every decade or so AI comes out of hibernation with a bang, dies with a whimper and then goes back into hiding for the next generation to re-discover; the proverbial "AI Winter". Improvements are made and things get incrementally better each time, but we're still a long way off from Lieutenant Commander Data and Star Trek the Next Generation. The things that machines can "learn" often result in trite conclusions or reproductions of simple algorithms that could have been done much more cheaply and with many fewer resources by trained humans. Is AI useful? Yes it can be, but is it going to replace human experts any time soon? Probably not. AI has gathered a neat bag of tricks over the decades, including the defeat of the best humans at Chess and now Go, but that's really mostly what they amount to; neat parlor tricks. The AI cheerleaders need to give it a rest before people once again grow weary of their big promises and equally big failure to deliver. The general public is too ignorant to know the history of AI, so they lap up the bullshit that's spoon feed to them, but those of us who've been around for a few decades in software know better. Sometimes I wonder if the "investors" writing the checks for AI startups have actually done their homework. Maybe P.T. Barnum was right when he said that there's a sucker born every minute.
If machine learning was all about "the right tools" then Tesla (and Google) would just Tensorflow their way towards a self-driving car...
(remember that when a vendor tries to sell your company a machine learning and/or analytics package...)
Yes, I'm sure Tesla/Google/etc., are using tools for self driving car, but the key challenges aren't the tools... and it's the same for most real world problems that haven't already been trivially solved via decision trees...
Microsoft is 'mainstreaming' malware and spyware and terrorist marketing tactics. Like any terrorist organization they need to be degraded and destroyed.
MS is desperately thrashing around after screwing the pooch on mobile.
The win in Go should not be brushed off. This was no brute force win, the game combinations exceed the number of atoms in the universe. The win demonstrates an AI next step up. With Go's downfall there is no "next game" left to conquer - that was it, the creme de la creme fell.
Is M$:
1. Landing on the FP several times a day?
2. Scratching their asses this very moment?
3. Live streaming exciting developer downtime?
4. YOUR POST HERE!
upload your munged data and drag and drop dozens of algos. chop out training sets and test your baby neural nets. come on.. Microsoft has been working hard to play nice and make amends. Forget your old playground grudges. Give yourself to a chance to try y out working ML algos even against your first Hadoops in the cloud.
But I am an avid reader of licensing agreements.
While the CLTK is mostly an MIT licensed library, most of its prerequisites are not.
The 1bitsgd library is not MIT licensed (it is either a no-modification maybe commercial license, or a non-commercial modification license.)
It requires Intel's mkl math library (or experimental support for openblas, which didn't help me get it to compile on linux!), and openmpi, with optional cuda support for GPU acceleration. I haven't figured out yet if it actually has opencl support or not, but hwloc's opencl/gl enumeration appears broken on gentoo with my current setup. So you either can use openmpi support cpu-only, or openmpi+cuda, but not opencl for both ATM.