Skip to main content

View Diary: NSA whistleblower Russell Tice on Countdown w/KO, Day Two (update X4) (152 comments)

Comment Preferences

  •  Then it's the wrong system (1+ / 0-)
    Recommended by:
    NCrissieB

    I come from a similar background, and have seen a small slice of what they're using.

    I'm pretty sure that it doesn't work quite the way you're envisioning it, but I don't know enough to be certain. However, if they really did design it that way, then the system needs to be scrapped, and a new one designed - and the spec must include working with a smaller dataset. It's a problem that has to be solved.

    In real life, a TARP is a cover-up.

    by mataliandy on Thu Jan 22, 2009 at 10:22:46 PM PST

    [ Parent ]

    •  How do you do that... (1+ / 0-)
      Recommended by:
      NCrissieB

      ... when if the range of the features of the training set is severely limited when compared to the range of the features to be detected when the system is applied to live data?

      Quick to judge, Quick to anger, Slow to understand; Ignorance and prejudice and fear walk hand in hand. -- Neil Peart

      by JRandomPoster on Thu Jan 22, 2009 at 10:39:13 PM PST

      [ Parent ]

      •  You write QA software (1+ / 0-)
        Recommended by:
        JRandomPoster

        That will generate a dataset you need for testing, perhaps pulling several dozen years' news stories from around the world, for example.

        But if it somehow simply can't be done - if it's not possible to use the technology without removing our rights, then it is the wrong tool to use. Scrap it.

        Technology does not get to decide what rights we have. Period.

        In real life, a TARP is a cover-up.

        by mataliandy on Thu Jan 22, 2009 at 11:03:42 PM PST

        [ Parent ]

        •  The problem of writing a driver... (1+ / 0-)
          Recommended by:
          blueoasis

          ... to generate sufficiently realistic test data is exactly the same of that of training a classifier without actually reading the communications used to train said classifier.

          You can't use the classifier you are building to generate the training set.

          And you can't build a test generator without training it with real data, either.  The patterns being sought are too statistically subtle to generate in your test set without the use of real world data.

          Quick to judge, Quick to anger, Slow to understand; Ignorance and prejudice and fear walk hand in hand. -- Neil Peart

          by JRandomPoster on Fri Jan 23, 2009 at 12:35:33 AM PST

          [ Parent ]

Subscribe or Donate to support Daily Kos.

  • Recommended (129)
  • Community (65)
  • Bernie Sanders (44)
  • Elections (40)
  • 2016 (38)
  • Climate Change (33)
  • Environment (32)
  • Culture (31)
  • Hillary Clinton (29)
  • Science (26)
  • Republicans (26)
  • Media (25)
  • Barack Obama (24)
  • Civil Rights (24)
  • Education (22)
  • Law (21)
  • Trans-Pacific Partnership (21)
  • Economy (19)
  • Congress (17)
  • Labor (16)
  • Click here for the mobile view of the site