I am reading Composing Music With Recurrent Neural Networks by Daniel D. Johnson. But I am really confused about the input passed to this network. If we pass notes of music along the time axis, then what is passed along the note axis?
The author says:
If we make a stack of identical recurrent neural networks, one for each output note, and give each one a local neighborhood (for example, one octave above and below) around the note as its input, then we have a system that is invariant in both time and notes: the network can work with relative inputs in both directions.
This might mean that the inputs passed to the network along the note axis are fixed representations of notes in the vocabulary. But I am not sure.
I am also having a hard time understanding the input passed to this network as the author explains a few paragraphs below. (Position, Pitchclass, Previous Vicinity, Previous Context, Beat).
Also, at some point, the author talks about RNN along the note axis. But in the architecture, there only seems to be RNN along the time axis. I would really appreciate is anyone could give me some more information to understand how this Biaxial Network is setup. This article by Deep Dark Learning was a little helpful but I am still not fully sure what is going on here.