Ian Goodfellow

With Yaroslav Bulatov, Julian Ibarz and others at GoogleGoogleGoogle is an American technology company founded by Larry Page and Sergey Brin and incorporated in 1998. It became an Alphabet subsidiary in 2015. Its businesses include internet search, advertising, software platforms, cloud computing, and artificial intelligence.Open the full entry, he submitted a paper on recognizing arbitrary multi-digit numbers from Street View imagery with deep convolutional neural networks, a sub-problem of recognizing multi-character text in natural photographs.⁠[1]

One night in 2014, while still a doctoral student, Goodfellow had the idea while celebrating a fellow student’s graduation at Les 3 Brasseurs, a Montreal bar; back home he coded the first example of what he named a “generative adversarial network.”⁠[2]

As first author with Yoshua Bengio, Aaron Courville and others, he submitted “Generative Adversarial Networks,” a framework for “estimating generative models via an adversarial process,” in which generator G is trained to maximize the probability that discriminator D makes a mistake.⁠[3]

With Jonathon Shlens and Christian Szegedy he submitted “Explaining and Harnessing Adversarial Examples,” examining why neural networks and other models consistently misclassify adversarial examples.⁠[4]

Goodfellow received his Ph.D. from the University of Montreal in 2014 and then worked at OpenAI and Google.⁠[5]

At Google Brain he did deep learning research and helped improve products; he then spent about a year at OpenAI before returning to Google as a research scientist for a little over two years.⁠[6]

Deep Learning, the textbook he co-wrote with Yoshua Bengio and Aaron Courville, was published by MIT Press in 2016, and its complete online version remains free to read.⁠[7]

He presented the tutorial on generative adversarial networks at NIPS 2016 and released a report summarizing it on December 31, 2016.⁠[8]

His list of presentations shows repeated talks on adversarial examples and adversarial training from 2015 to 2019, including a live demo of physical adversarial examples at GeekPwn 2016 and a “35 under 35” talk at EmTech 2017.⁠[9]

MIT Technology Review named Goodfellow, then 31, one of its Innovators Under 35 for inventing a way for neural networks to improve by working together; he was then a staff research scientist on the Google Brain team.⁠[10]

He was then heading a machine learning security team at Google, saying: “hopefully we can make significant advances in security before we’re too far in.”⁠[2]

Goodfellow joined Apple in March 2019 as a director of machine learning in the Special Projects Group; at Google he had worked on GANs and security, including adversarial attacks.⁠[5]

He appeared on the Lex FridmanLex FridmanLex Fridman is a Soviet-born, US-raised computer scientist and podcast host who works as an MIT research scientist on human-AI interaction and autonomous driving. His Lex Fridman Podcast is known for hours-long conversations with guests including heads of state and technology leaders.Open the full entry Podcast as the author of the textbook Deep Learning and the person who coined the term generative adversarial networks, in an episode devoted to GANs.⁠[11]

Goodfellow left Apple over its in-person work policy and was subsequently set to join Alphabet’s DeepMind; DeepMind declined to comment at the time.⁠[12]

Goodfellow officially announced that he was joining DeepMind as a research scientist on its Deep Learning team.⁠[6]

His Google Scholar page lists his affiliation as DeepMind with 443,694 total citations; his most-cited works are the GAN paper (2014), Deep Learning (2016), the adversarial examples paper (2014), TensorFlow (2015) and “Intriguing properties of neural networks” (2013).⁠[13]

Personal website: Public link: https://www.iangoodfellow.com/ (opens in a new window)⁠[14]

DeepMind: Official website: https://deepmind.google/ (opens in a new window)⁠[13]