Re:Court rules drug dog barely more accurate than a coin fli...
by saint_fiasco on Friday August 07, 2015 @01:27PM
(#10023692)
Humans don't properly implement AdaBoost on their heads.
If a dog alerts an officer, confirmation bias will cause the officer to think "now that I look at him closely, he sure seems to be acting strange..."
For bad implementations of adaptive boosting, weak information can be worse than no information.
Re:Court rules drug dog barely more accurate than a coin fli...
by shkkmo on Friday August 07, 2015 @02:29PM
(#10024101)
But if you are going to use that reasoning, then you need to remove the dog from the equation and only judge the officers on their actual success rate, utilizing any method at their disposal. What should this rate be? 50%? Any officer who falls below that rate should be forced to get approval from another officer with a higher success rate? They can then choose to use dogs to supplement certain officer's accuracy without giving them incentive to train the dogs to be "probable cause generators".
Using a weak classifier to serve as the basis for getting around a constitutional right is pretty messed up.
Re:Court rules drug dog barely more accurate than a coin fli...
by daenz on Friday August 07, 2015 @03:52PM
(#10024606)
> Using a weak classifier to serve as the basis for getting around a constitutional right is pretty messed up.
I was playing the devil's advocate, in case you missed that part, and my point was that a classifier being weak does not necessarily make it bad, as evidenced by AdaBoost.
To play the devil's advocate, if the dog alerting to drugs is only one of the classifiers in play (others being the suspects look/act under the influence), and if it's weighed correctly in an officer's mind, it is a key part of the strong classifier.