Fix optimization not always registered and python 3 crash

Moved general FAQ to top-level index page.

[BUG] Important bug fix in an optimization.

[WIP] Fast compile gpu

Propagate shape information to ConvOp gradients

Abergeron dnn

Open Machine Learning Workshop 2014 presentation

Reduction(axis!=0) of joins(axis=0) wasn't working on the right axis.

fix gh-2109

Added to test to determine the behavior of the BLAS and intialize the me...

keep_dims -> keepdims in theano.tensor.nnet documentation

I'm not able to reproduce the problem. It happen in the DLT
logistic_sgd with fast_compile optimizer.

Added a new decorator and tagged a few opts for optimization when using the gpu or optimizer=fast_compile.

Register to the gpu tag more opt, as we don't have all the other case on the GPU otherwise in fast_compile.

fast_compile.