DokuWiki

The following are the most likely applications of neural networks in GridMusic doable in Summer 2015:

• Net for harmonizing a given melody (using a class derived from Sequence)
• Net for generating a seed melody (adding a button to the Gen section of the UI)
• Possibly including an analysis/classification unit for supporting the AI class

In theory, a neural network can be trained to generate an expansion sequence in different styles simply by training it beforehand with instances of that style. This implies that the network is trained prior to being used in GridMusic, which means it must be a separate process. The easiest way to do this (without switching over to Clojure or something):

• Develop a base network class the represents the complete state of network in training or generation mode
• Write a program for training an instance of the network which it then saves as a file
• Write a class that reads a given file containing the description of a network and runs it on demand

Of these, the first and last will be part of GridMusic. The second will be a separate sketch.

Query: should the base network class be a Processing library?