Ian’ GANfather’ Goodfellow, the inventor of generative adversarial network recently left Apple to join top AI research lab – DeepMind.

One of the best talents in the field of AI, Goodfellow has always been on the top of his game – his research work has been cited over 1.9 lakh times. Some of his popular work includes generative adversarial nets (cited 46,910 times) and research on deep learning (cited 42135 times).

DeepMind has been focused and been in the limelight for projects like AlphaGo, a program that beat the world champion at Go in a five-game match, and AlphaFold, which found a solution to a 50-year-old grand challenge in biology. At DeepMind, Goodfellow is working with Oriol Vinyals’ (principal research scientist) team. The duo has collaborated earlier on research projects. In 2016, they worked on a project on the TensorFlow interface and its implementation during their time at Google. Previously in 2015, the two worked together on a neural network optimization problems research project. 

At DeepMind, Vinyals has been working on AI, with particular emphasis on machine learning, deep learning and reinforcement learning. With Goodfellow bringing his GAN capabilities to the DeepMind team, the best of both worlds have joined hands. Two stalwarts of the field coming together opens up a new world of opportunities.

Ian, the good fellow 

Goodfellow headed the special project group (SPG) at Apple, which works on cutting-edge tech innovations and futuristic products. The SPG reportedly houses Apple’s self-driving car division, Project Titan, led by Tesla’s former Senior Engineering VP, Doug Field.

Goodfellow had interviewed at Facebook in the fall of 2013, listening to Mark Zuckerberg philosophize about DeepMind as they strolled across the campus courtyard. Then he’d turned Zuckerberg down in favour of a job with Google Brain.

The Stanford graduate got associated with Google in 2013 as an intern. He worked with a team that created a deep neural network to read address numbers from Street View imagery. The system was used to add or update the location of over 100 million houses within the first few months of its deployment.

Interestingly, Goodfellow left Google twice, once to join OpenAI and the second time to join Apple.

Between 2015-16, as a member of the Google Brain team, he worked on deep learning, both in terms of basic research and in terms of improving products. Furthermore, he curated a textbook on deep learning along with his PhD thesis advisors. His research is widely cited in academic articles about AI.

Later, during his stint at OpenAI, the Elon Musk and Microsoft-funded AGI think tank, he was a crucial part of the research team for generative models and adversarial examples.

Ian GANfather Goodfellow

“If an AI can imagine the world in realistic detail – learn how to imagine realistic images and realistic sounds – this encourages the Al to learn about the structure of the world that exists,” Goodfellow said in the book Genius Makers by Cade Matz. “It can help the AI understand the images it sees or sounds it hears.”

Goodfellow’s greatest claim to scientific fame is as a part of the team that invented the generative adversarial network (GAN). “If it hadn’t worked, I might have given up on the idea.” In the paper Goodfellow published on the idea, he called them “generative adversarial networks,” or GANs. Across the worldwide community of AI researchers, he became ‘the GANfather.’

Yann LeCun called GANs ‘the coolest idea in deep learning in the last twenty years.’ When Geoff Hinton heard this, he pretended to count backwards through the years to make sure GANs weren’t any cooler than backpropagation before acknowledging that LeCun’s claim wasn’t far from the truth.

The WFH debate

As communities rush to attain normalcy, the WFH debate has again popped up. One of the instances that triggered the conversation was the Apple machine learning head’s resignation.

Famously known as the man who has ‘given machines the gift of imagination‘, the inventor of generative adversarial networks (GANs), he left Apple for DeepMind, citing the former’s lack of flexibility in work policies. This was in the backdrop of Apple calling back its employees to work from its offices.

In an email to fellow Apple employees, Goodfellow said he strongly believes more flexibility would have been the best policy for his team.