Yoshua Bengio

Overview

Yoshua Bengio is a Canadian computer scientist, born on 5 March 1964 in Paris.^[2] He has been a full professor of computer science at the Université de Montréal since 1993.^[2] In 2018 he shared the ACM A.M. Turing Award with Geoffrey Hinton and Yann LeCun, for breakthroughs that made deep neural networks a critical part of computing.^[3]

Education and early research

Bengio studied at McGill University, where he earned a bachelor's degree in computer engineering and a master's and PhD in computer science. His PhD, completed in 1991, dealt with neural networks and sequence recognition. He then did postdoctoral work at MIT with Michael I. Jordan and at AT&T Bell Labs.^[2]

In 1994 he published, with Patrice Simard and Paolo Frasconi, the paper Learning long-term dependencies with gradient descent is difficult, in IEEE Transactions on Neural Networks.^[5] The title names the problem. When a network must connect information across long spans of a sequence, the training signal tends to fade or become unstable, which makes learning those links hard.

In the years after his doctorate he combined neural networks with probabilistic sequence models such as hidden Markov models, which describe a sequence as moving through hidden states that produce what we observe. This combined work was used in an AT&T/NCR system for reading handwritten checks.^[3]

Major contributions

In 2003 Bengio, Réjean Ducharme, Pascal Vincent and Christian Jauvin published a neural probabilistic language model in the Journal of Machine Learning Research (JMLR). The model learned word embeddings. A word embedding represents each word as a list of numbers, arranged so that words used in similar ways receive similar lists. ACM credits this work with introducing high-dimensional word embeddings.^[3]

In 2014 his group, with Dzmitry Bahdanau and Kyunghyun Cho, introduced the attention mechanism for neural machine translation. Attention lets a model look back at the words of the source sentence that matter for each output word, instead of squeezing the whole sentence into one fixed summary.^[2]

In the same year he co-authored the 2014 paper on generative adversarial networks (GANs) with Ian Goodfellow and others. In a GAN, a generator tries to produce realistic examples while a discriminator tries to tell them apart from real ones. In 2016 he published the textbook Deep Learning with Goodfellow and Aaron Courville.^[2]

Career

Bengio founded Mila, the Quebec AI Institute, and was its scientific director until 2025. He is now its founder and scientific advisor. He holds a Canada CIFAR AI Chair.^[1] He chairs the International AI Safety Report, whose full report was published in January 2025. In June 2025 he launched the nonprofit LawZero, where he is co-president and scientific director.^[1]

Honours and influence

His other honours include Officer of the Order of Canada (2017), Fellow of the Royal Society (2020) and the Princess of Asturias Award (2022). In 2025 he shared the Queen Elizabeth Prize for Engineering with Geoffrey Hinton, John Hopfield, Yann LeCun, Jensen Huang, Bill Dally and Fei-Fei Li.^[4] The Turing Award carries a prize of one million dollars, funded by Google.^[3] The word-embedding and attention ideas from his group are now common components of language-model research.