Finally moving on to interesting stuff with the nets. From wave-form encoding and into actual sound. That is, changes through time instead of static frames - first task will be learning to play the drums. I've gathered an hour and 17 minutes of unaccompanied kit-drum recordings, and hope to have an RNN training on it by the end of the night. #neuralnets #RNN #music #theano #lasagne #machinelearning
Fox is state media now. Rupert #murdoch is Minister of Information [sic]. https://twitter.com/realDonaldTrump/status/884931352201756672
Still working on a good audio frame encoder. I've got one net trained and set aside that I'm not satisfied with and am trying a second pass which will take a few more days. No weight decay or regularization of activations on the encoding this time, instead the learned vector will be rescaled to produce a new target encoding for a next generation for multiple cycles. #lasagne #theano #machinelearning #audio #neuralnets
[Old] Locking #Science Open with Decentralized Scientific Archives https://blog.datproject.org/2017/06/21/locking-science-open/
Waiting on weights, waiting on weights, waiting on weights to train. Looks like a few more days, maybe a week, and I'll have satisfactory neural encoder/decoder for audio frames and can start playing with RNNs using it to start learning instruments, notes, & rhythm, along with other tasks to include speech recognition and speaker re-identification. #theano #lasagne #neuralnets #machinelearning #audio
