University of Hong Kong students have designed and trained neural networks that can correctly identify the genre of a song, based on features such as tempo and harmony, with greater than 87 percent accuracy — considered a very hard problem in computer science.