MIT researchers machine learning used to test systems data set put it under the spotlight. The team led by computer scientists is 3.4 percenthas been labeled incorrectly or incorrectly and uses these data sets. artificial intelligence They found that they could cause problems with their (AI) systems.
See Also. “Razer begins production of smart face masks with RGB lighting”
You may be interested in
Apple buys 25 AI companies in five years
Each one 100.000Among the datasets that have been quoted more than once, newsgroups such as Amazon and IMDb text based There are also those. Amazon product reviews, actually positive Although negative Errors such as mislabeling as (or vice versa) stand out.
If image-based errors are animal species It comes out by mixing. In addition, the photos are in the middle of the less conspicuous object mislabeling also mentioned. For example instead of mountain bikes water bottle Using the tag.
It can cause serious problems
In YouTube videos voices Although another dataset based on the image of a YouTuber speaking to the camera for three and a half minutes could only be heard in the last 30 seconds “church belllabels it as “. Bruce Springsteen’s performance”orchestraThe classification as “also draws attention as another mistake.
Artificial intelligence and machine learning, many of them public using data from a subset of data sets come to a conclusion was built to. Bad tags for machine learning systems to serious troubles may cause. Researchers have also set up a website so everyone can keep an eye on these errors.