Modern Deep Learning & Transformers

Fei-Fei Li

Built ImageNet, the large labeled dataset that made the 2012 deep learning breakthrough in computer vision possible in the first place.

Fei-Fei Li (b. 1976) built the dataset that made modern computer vision possible before anyone had the architecture to fully exploit it. Starting in 2006, she led the construction of ImageNet: over 14 million images, hand-labeled across more than 20,000 categories using crowdsourced annotation at a scale nobody in computer vision had attempted before. The premise was that progress in vision wasn't bottlenecked purely on algorithms — it was bottlenecked on data, and nobody had built a dataset large and clean enough to prove it.

She was right. The annual ImageNet competition became the field's benchmark, and in 2012 Geoffrey Hinton's team entered AlexNet, a CNN that beat the next-best entry by a margin large enough to end the argument over whether deep learning could outperform hand-engineered vision features. Without ImageNet's scale, AlexNet would have had nothing to prove that result on — the dataset and the architecture were both necessary, and neither gets there alone.

Li went on to co-direct the Stanford Human-Centered AI Institute and has been a consistent voice for keeping human impact and ethics central to AI research, distinct from the pure capabilities race much of the rest of the field is running.

See also: Geoffrey Hinton

Learn more: Computer Vision · Wikipedia: Fei-Fei Li

Mentioned in

Lessons where this comes up in context.