Re:Human psychology and behavioral studies overlook 85 perce...
by bjourne on Thursday February 07, 2019 @07:18AM
(#19104731)
> Years after pondering this article, I felt I had figured it out: psychology is secretly the study of particular subset of 18 to 21 year old American women.
There are many analogs of that in Computer science too. For example in computer vision, CIFAR-10 is a de facto standard for measuring performance. A set of 60k 32x32 images. Good results on that data set doesn't necessarily translate well into real-world performance. But what can you do? Gathering huge data sets and having humans annotate them is incredibly expensive.
Another ml example is music recognition. There are several state-of-the-art methods for detecting notes in music. So one Chinese researcher tried to apply them for Jingju music https://www.youtube.com/watch?v=NJsTl342RhI. The results were... less than stellar.
There are many analogs of that in Computer science too. For example in computer vision, CIFAR-10 is a de facto standard for measuring performance. A set of 60k 32x32 images. Good results on that data set doesn't necessarily translate well into real-world performance. But what can you do? Gathering huge data sets and having humans annotate them is incredibly expensive.
Another ml example is music recognition. There are several state-of-the-art methods for detecting notes in music. So one Chinese researcher tried to apply them for Jingju music https://www.youtube.com/watch?v=NJsTl342RhI. The results were... less than stellar.