Jazz Deep Learning at HMC

A group picture of the Summer 2016 ImproVisor research team
A group picture of the Summer 2016 ImproVisor research team

At Harvey Mudd College (during the summer of 2016), our team worked to implement techniques for generating improvised Jazz music using neural networks. The team worked on several projects, whose information and results should be available online sometime in the near future. The product I created alongside my team member Daniel Johnson is a recurrent auto encoder network which can be trained to recognize and reconstruct musical features of sheet music.

The trained recurrent auto encoder is able to interpolate between musical phrases.
The trained recurrent auto encoder is able to interpolate between musical phrases.

Our work was incorporated into the main branch of Dr. Keller’s Impro-Visor software (github.com).

My work on the Java implementation of the network (as well as its use in musical trading) is located in the lstmprovisor-java repository.

Daniel’s work on the training utility is located in the lstmprovisor-python repository.

Thanks to Dr. Robert Keller, Harvey Mudd College, the National Science Foundation, my teammates Daniel Johnson, Sam Goree, Mackenzie Kong-Sivert, Wenbo Cao, and our interns Mira Jambusaria and Joshua Zhao for the excellent summer.