Université de Montréal

A French-language university in Montreal

Université de Montréal is a French-language public research university in Montreal, Quebec. Its official website lists 16 faculties and schools, including Arts and Science, Medicine, Law and Music, and names HEC Montréal and Polytechnique Montréal as its affiliated schools.^[8]

From branch campus to autonomy

The university was founded in 1878 as a branch of Université Laval. It became a university in its own right when a papal charter granted it full autonomy on 8 May 1919.^[1] That date, rather than the founding date, marks the point from which the institution governed itself instead of answering to another university.

Deep learning

Deep learning is machine learning built on neural networks that stack many layers between input and output. Each layer transforms the data, so the network can build up increasingly abstract representations instead of relying on features designed by hand. The university's modern reputation rests largely on this field. Yoshua Bengio has been a faculty member in computer science at the university since 1993,^[2] and Mila, the deep-learning institute he founded in 1993, has been run in partnership with McGill University since 2017; it is described as the world's largest academic research centre for deep learning.^[1] Having an institute of that size attached to the university means the field is represented there by a broad cluster of research groups rather than a single laboratory.

Papers written at the university

Several papers that shaped modern language technology were written there. "A Neural Probabilistic Language Model", published in the Journal of Machine Learning Research in 2003, was written by Bengio with Réjean Ducharme, Pascal Vincent and Christian Jauvin of the Université de Montréal, and it introduced learned word feature vectors for language modelling.^[3] A word feature vector is a list of numbers standing in for a word, learned from text rather than written by hand, so that words appearing in similar contexts end up with similar numbers; the model can then generalise across words it has seen in comparable use instead of treating every word as an isolated symbol.

In 2014 the "Generative Adversarial Networks" paper by Ian Goodfellow and colleagues, including Bengio, was written at the university.^[4] A generative adversarial network trains two models against each other: one produces samples, the other judges whether a sample is real or generated, and that competition pushes the generator towards more convincing output.

Also in 2014, the attention-based neural machine translation paper by Dzmitry Bahdanau, Kyunghyun Cho and Bengio appeared, listing Cho and Bengio at the Université de Montréal.^[5] Attention lets a translation model look back at different parts of the source sentence while producing each word, instead of compressing the whole sentence into one fixed-length vector first.

People also passed through. Tomáš Mikolov, later the lead of word2vec, spent time in Bengio's lab at the university during his PhD.^[6]

Recognition and position

Bengio shared the 2018 ACM A.M. Turing Award with Geoffrey Hinton and Yann LeCun.^[7] The award is given by the Association for Computing Machinery, which is what the "ACM" in its name stands for. For the university, the combination is unusual: a broad, century-old faculty structure on one side, and on the other a dense concentration of deep-learning work that produced foundational papers in language modelling, generative models and machine translation, plus an institute that trains and hosts researchers in the field.