3 ms·
Yeah trying to make a RNN learn about simple (but "complicated") priors is an exercise in how not to do deep learning. Just think how big a model should be to
by raverbashing 8y ago
Yeah trying to make a RNN learn about simple (but "complicated") priors is an exercise in how not to do deep learning.
Just think how big a model should be to understand self similarity in 2 different parts of a song.
- AstralStorm 8y agoUnfortunately a bigger more complex RNN learns the songs by heart and doesn't generalise. Beta-VAE units seem essential and they're not doing well either.