History
What defunct deep learning library was started in 2007 by the Montreal lab now known as Mila?
Theano
Proposed by @GokuMohandas
Card 1 of 259. Answer: Theano
Theano
Proposed by @GokuMohandas
Card 1 of 259. Answer: Theano
About · Quizzes · History of AI · To suggest new questions or report an error send me a dm.
Theano. It was developed from 2007 in Yoshua Bengio's lab at the Université de Montréal, then called LISA (Laboratoire d'informatique des systèmes adaptatifs) and only later renamed the Montreal Institute for Learning Algorithms (MILA), now Mila. Strictly, Theano was a Python compiler for math on multi-dimensional arrays, on which deep learning libraries were built.
You defined an expression symbolically; Theano derived its gradients automatically, optimized the graph and generated C++ or CUDA code for a CPU or GPU. James Bergstra, Olivier Breuleux, Frédéric Bastien and colleagues presented it at the SciPy 2010 conference. Keras first ran on it, as did Lasagne, Blocks and PyMC3.
On 28 September 2017 Bengio announced that MILA would stop developing Theano after version 1.0, noting that strong industry-backed frameworks now existed. Theano 1.0.0 came out on 15 November 2017. The PyMC developers took over maintenance, and the code lives on in their forks, Aesara and then PyTensor. Theano is an early example of the static, build-then-run graph style that TensorFlow 1.x also used.