/LearnedNormPooling

Attempt at an implementation of http://arxiv.org/pdf/1311.1780v7.pdf in Theano/Lasagne

Primary LanguagePython

LearnedNormPooling

Attempt at an implementation of http://arxiv.org/pdf/1311.1780v7.pdf in Theano/Lasagne.

How to run

Make sure you have Lasgane and Theano properly installed. Then:

python mnist_test.py

TODO

Scan function/symbolic tensor functions are not optimized! Takes way too long on a CPU. Next step, try one of the following:

  • Refactor code to work in GPU, test runtime
  • Consider alternative implementation that overrides the Pool and PoolGrad op classes.