Mood - Meeting 1

I. Notes from meeting with Professor Winston and Sajit Rao.

• Expectation of what to listen to with patterns of attentional state -> still perceive even if fragmented / other cord

Sajit Advise:

• Show in our project why attentional state useful / better than, say, neural nets
• Find invariant about Jazz / Classical in changes of attentional state that sticks out.
• Map operations in vision to operations in sound - what is selection?
• Think hardware support -> What in our brain could be a primitive for sound attention?

II. April 4th meeting:

Focus project

• We only have 1 month ` get a feasible project
• single instrument = piano single handed
• maybe instead input = midi file
• Project could be: Play 20 notes and let the program figure out what type of music it is.

Experiment with

• attentional state
• neural net learning
• arch learning

What to base our system inputs on:

• notes (pitch)
• difference in notes (d(pitch) / dt)
• past 3 notes / seconds, for increasing values of 3 -> that is, experiment with different sizes of past window
• volume of note (how strong)
• d(volume) / dt
• speed of notes
• d(speed) / dt ` how the speed of playing changes with time

What to come up with

• Weighting of music will be a 4-vector (ex: [20, 0, 0, 2] on the [Jazz, Classical, Romantic, Modern] scale)

III. Questions:

• How to train the program
• How does it learn the first pattern
• How can it do multiple learning simultaneously