How widely is Theano used in deep learning research?

Is Theano a good start to learn the implementation of machine learning algorithms?

Will learning the implementation of something like a feed forward network really help? Do graduate students implement neural networks or other algorithms at least once during their college days?


I have a reasonable idea about feed forward and recurrent networks, backpropagation, the general pipeline for a machine learning problem and the necessary mathematics.


Theano used to be very popular in the last few years. However, from seeing what the top research labs in the world currently use instead of theano (Facebook AI and Google Deepmind), I would either go for caffe or Torch7, with a stronger preference on the latter.

The reason is that you can implement from Recurrent Neural Nets to Convolutional Neural Networks very easily and there is a wide range of examples on github (https://github.com/torch/demos).

Graduate students in NYU, Stanford, Oxford definitely implement advanced models in their studies; your background will help you a lot, a great place to start and familiarise with such models, and implement them in Torch7, is the lessons of Prof. Nando de Freitas https://www.youtube.com/user/ProfNandoDF.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy