This proposal will add more convolution operations to aid the users rather than implementing these operations on their own.These operations would be a great addition to not only Theano but also the libraries that are built using theano like Keras and Lasagne.This project will also implement a meta-optimizer on the new gpu-backend which will choose the fastest convolutions from the many available, thus running these operations as fast as possible.

Student

Mohammed Affan

Mentors

  • Frédéric Bastien
  • Arnaud Bergeron
  • Gijs van Tulder
  • Pascal Lamblin
close

2017