Slashdot Mirror

← Back to Stories (view on slashdot.org)

How Machine Learning Ate Microsoft

Posted by samzenpus on from the doing-it-smarter dept.

snydeq writes Yesterday's announcement of Azure Machine Learning offers the latest sign of Microsoft's deep machine learning expertise — now available to developers everywhere, InfoWorld reports. "Machine learning has infiltrated Microsoft products from Bing to Office to Windows 8 to Xbox games. Its flashiest vehicle may be the futuristic Skype Translator, which handles two-way voice conversations in different languages. Now, with machine learning available on the Azure cloud, developers can build learning capabilities into their own applications: recommendations, sentiment analysis, fraud detection, fault prediction, and more. The idea of the new Azure offering is to democratize machine learning, so you no longer need to hire someone with a doctorate to use a machine learning algorithm."

21 of 96 comments (clear)

  1. Autocorrect :) by Anonymous Coward · · Score: 5, Funny

    Ate or Aid?

  2. How Bing learns by lu-darp · · Score: 3, Interesting

    So how did MS apply "machine learning" to make Bing not suck? By holding an internal competition to see who's algorithm processed "user improvement program" data best. So that essentially meant training it up to match Google search results (and presumably, which links "consenting" users clicked).

    (OK, I'm sure they've come a long way since then on their own merits, but we can't let them live that one down ;-)

    1. Re:How Bing learns by brunes69 · · Score: 4, Insightful

      The problem with your statement is you make the assertion that Bing no longer sucks, which is false. Bing is still horrible.

    2. Re:How Bing learns by Marginal+Coward · · Score: 2

      I dunno...I just put the following query into Bing: "is bing horrible", and it came up with Why Bing Sucks. Top 5 Reasons. So...it sucks, but it certainly isn't horrible.

    3. Re:How Bing learns by Marginal+Coward · · Score: 4, Interesting

      Although the above was just a joke, I actually clicked on the link after I submitted it, and it turns out to be an old page from 2009. It provides the follows searches which it says "just don't work" on Bing (in 2009):

      “Was Einstein married?”
      “What did Benjamin Franklin invent?”
      “What is the top selling album of all time?”

      I did a quick comparison of those three between Bing and Google, and the results seemed pretty comparable. In fact, I thought Bing did a little better on the first two, and Google did a little better on the last one - primarily because it provided a nice blurb from Wikipedia in the results.

      So, although I think we can all agree that Bing was "horrible" in the past, it's come a long ways. It's not like in the old days when Google was clearly the best - I think you could use any of the major search engines now and do just fine.

    4. Re:How Bing learns by lgw · · Score: 3, Interesting

      DuckDuckGo uses Bing, but it does some nice enhancements, like putting Wikipedia results first very often. If you want to live a Google-free life, and don't want to give your search history to MS either - ddg.gg.

      --
      Socialism: a lie told by totalitarians and believed by fools.
    5. Re:How Bing learns by doom · · Score: 2

      Also: blekko.com, startpage.com. There's no particular need to use google for web searches any more.

  3. Dammit! by nospam007 · · Score: 4, Funny

    "recommendations, sentiment analysis, fraud detection, fault prediction, and more."

    I'll never get a job again if they use that in interviews.

  4. Youpie I just wasted 5 years of my life.... by ProzakLord · · Score: 3, Interesting

    Baing about to finish a PhD this is worrying thinking that the deep understanding of a technique can be replaced by a well programmed API. But of course managerial people, a.k.a decission makers will eat that raw. If you want my data just ask for it.

    1. Re: Youpie I just wasted 5 years of my life.... by elcano · · Score: 2

      Sure. With the new API they will know what kernel to apply to their SVM learner, and the system will set the correct data types and preprocess the data just in the way that a parametric model needs, and will know when to segment your data when it is appropriate to apply different models to each segment, and will consider misclassification costs, and will know how to handle unbalanced datasets, and how to prevent overfitting, and when it is ok to apply ensemble models, and will write a summary justifying the assumptions made during the modeling process. Sure. What this will cause is a huge track recird of failed projects and a bad name for machine learning as a discipline. Very sad for qualified practitioners.

  5. Correlation and causation again by louic · · Score: 3, Insightful

    I foresee a bright future where lots of correlations will be found. Without "a PhD" or someone who knows what they are doing ALL software is worthless.

    1. Re:Correlation and causation again by gruntkowski · · Score: 5, Insightful

      That is so correct. I've tested the machine learning software on Azure. It is very nice and quite powerful. But without knowing what you are really doing, you probably get results which seem nice, but are in fact complete bogus. If you do not know what overfitting is for example, good luck using machine learning algorithms. If some manager starts using this, may god have mercy on us all...

    2. Re:Correlation and causation again by Dr_Barnowl · · Score: 3, Interesting

      Sounds like Reason..

      Reason allows users to specify in advance the decision they want it to reach, and only then to input all the facts. The program's task was to construct a plausible series of logical-sounding steps to connect the premises with the conclusion. The only copy was sold to the US Government for an undisclosed fee.

  6. Microsoft Bob by sycodon · · Score: 4, Funny

    Bob set machine learning back a century.

    --
    When Fascism comes to America, it will call itself Anti-Fascism, and tell you to give up your guns.
  7. Re:Call me paraniod, but ... by Lennie · · Score: 2

    I doubt it. They are in the business of selling products and services, they don't care what they can sell. They are a business trying to make money and stay relevant.

    If running a porn streaming service wouldn't damage their image and was something they thought they knew how to run well and make good money on, I'm sure they would just add it to their list of services.

    Now to be a bit more specific, of course they want your data. You see this happening especially on the consumer side.

    For example: where can I get a copy of SkyDrive/OneDrive/whatever which I can run on my own systems ?

    Anyway, I can't use Azure, I'm a foreigner:
    http://media.ccc.de/browse/con...

    --
    New things are always on the horizon
  8. Re:Call me paraniod, but ... by jbolden · · Score: 4, Informative

    For example: where can I get a copy of SkyDrive/OneDrive/whatever which I can run on my own systems ?

    SharePoint will do that.
    For that matter you can run the entire Azure suite in your private location: http://www.microsoft.com/en-us...

  9. Re:Call me paraniod, but ... by Lennie · · Score: 4, Interesting

    Let me be clear: what applies to Azure as a foreigner applies also to Amazon/AWS, Google, Rackspace, IBM/SoftLayer, CenturyLink, DigitalOcean, Vultr, Linode, PeerOne or any other US-based company (even if they run the service in Europe for example).

    But I noticed there are others in the world, for example on the OpenStack Marketplace:
    http://www.openstack.org/marke...

    --
    New things are always on the horizon
  10. Does Bing suck? by jbolden · · Score: 2

    Did Bing suck. I did the Bing vs. Google head to head test about 18 month back a few times. (looks like it might still be online at: http://www.bingiton.com/ And I most typically scored 3-Google, 2-Bing with often Bing having some interesting results Google didn't have. For example Bing tends to do better in hitting a better diversity of current information. Bing may be a bit behind possibly and I'm not even comfortable saying that, but sucks no.

    1. Re:Does Bing suck? by Anonymous Coward · · Score: 2, Interesting

      To me it seems more like Google has got _worse_ than Bing has got better.

      Too often when I put in very specific keywords that very relevant to a particular article Google doesn't show the most likely hit, it instead tries to suggest what Joe Six pack might want and that I actually made typos to the keywords or typed those keywords by mistake and ignores some of them.

      Then when I turn on Verbatim on Google, it gives me a list of all the pages that contain those keywords in a useless order (which could include link spam sites, or similar crap at the first page). Doesn't seem to be ordered by page rank or importance.

      Bing doesn't do better either. So it seems more like Google has got worse than Bing has got better.

      When I finally find the article (by trying other words or narrowing down by site), yes my original keywords are there.

  11. Re:no longer need to hire someone with a doctorate by Trepidity · · Score: 2

    More likely, there will be a basic set of functionality that can be used by Mr Below Average coder to generate a bunch of spurious correlations.

    I don't think getting the machine learning to "work" is going to be the hard part, in the literal sense of the code running and generating stuff. But if you have no understanding of statistics, the conclusions you draw are likely to be invalid.

    --
    10 PRINT CHR$(205.5+RND(1)); : GOTO 10
  12. Re:Call me paraniod, but ... by Lennie · · Score: 2

    But important parts are missing.

    Some examples:
    - AzureAD, specifically ACS
    - Site Recovery for disaster recovery

    These are all online services with no buy/download equivalent from Microsoft.

    --
    New things are always on the horizon