Hello Lines ///////
… I would like to dig deeper into all this deep learning topic. I am musician with some max8 knowledge. I did a Music Hackspace workshop on back-propagation, agents, weights and simple networks. But now I would like to get my head around in a more efficent way because max is so slow doing this kind of stuff…
Please recommend Books worth reading on AI, Neural Network, DL, ML to me…
All depends on what you are trying to do!
In general though, I’d have thought it’s worth doing some basic reading on multivariate analysis, non-linear optimisation and Bayesian analysis etc to get some grounding.
Data analysis techniques (eg self-organising networks etc) is really helpful if you are planning on training up different algorithms, especially with anything other than a trivially simple data set.
If you are developing your own stuff then reasonably well-written code is vital if you are doing anything complicated (having an understanding of fundamental algorithms eg Numbeical Recipes, Knuth Art of Computer Programming etc is really vital). Look at writing the final code in eg C or C++
Prototyping and developing algorithms can be more efficient in something like Matlab or Mathematica whilst you are still playing with ideas. (APL is also very useful but a bit out of fashion - but is excellent for abstracting these sort of algorithms)
I’d then look at actual ideas for implementing artificial neural networks, radial basis functions, Markovian and Bayesian stuff etc etc…
Usually the hardest thing is getting data sets into a state where you can analyse them and interpolate (ie train an ANN)
Just my thoughts - take it with a pinch of salt though - if three people post with suggestions you’ll likely get four different opinions !
I want to train a network with a range of chord types, e.g. drop2, drop3, drop 2+4, 3- and 4-part quartal voicings for major, melodic minor, harmonic minor modes… and clusters and their permutations in all keys.
I’ve thought of using midi data, but I don’t know how to store it effectively so that I can use my library to generate specific tasks without starting the training from scratch each time…
As I said, I’m just starting with this stuff
In order to do what with it though ? Are you looking at generating chord progressions ? From what?
Data representation, management and processing is always the hardest part of any task! The rest is always easier…
Maybe look into Markov chains - a typical ANN may not be what you are looking for.
I’d suggest defining exactly the problem you are trying to solve - try writing it out and thinking about the edge cases - that’ll make it easier to decide how to approach it…
Good luck !