MMOOZZGG.com

Glossary

Who is Fei-Fei Li?

In short

Fei-Fei Li is a professor of computer science at Stanford, the creator of the ImageNet dataset and the founding director of the Stanford HAI institute.

People · Fei-Fei Li

MMOOZZGG.com editorsChecked: 1 min read

For a child

Imagine a huge album with millions of photos: cats, cars, flowers, guitars, and under each one a label saying what it shows. Fei-Fei Li came up with the idea of building such an album for computers. Before, they were shown few pictures, and they got confused. When programs got so many examples, they learned to recognize objects in photos almost as surely as you recognize your friends in class.

In plain words

Li is a professor of computer science at Stanford University. Her main work is ImageNet, a huge set of photos where people labeled by hand what each one shows. In the 2009 paper, the set already held 3.2 million images in 5,247 categories, and the goal was 500–1,000 pictures for each concept. The labels were collected through Amazon Mechanical Turk, a service where thousands of people do small tasks. Why it matters: a neural network learns from examples, and the more there are, the more accurately it recognizes objects. An app that tells a dog’s breed from a photo rests on the same idea.

For business

One lesson from Li’s work matters most for business: the quality of a model depends on data, not only on the algorithm. ImageNet showed that a large, carefully labeled set of examples gives a jump in accuracy. In a company, this means spending on collecting and labeling data: photos of defects on a conveyor, document scans, pictures of goods on a shelf. The price depends on the number of examples, the difficulty of labeling and who does it, your own staff or a contractor. The main risk is labeling errors: the model will repeat them in its work. Before launch, check whether you have enough of your own data, who owns the rights to the images and whether they contain personal data. What to measure and how to collect such data is covered in the section on data for AI.

For a specialist

The paper “ImageNet: A Large-Scale Hierarchical Image Database” (Deng, Dong, Socher, Li, Li, Fei-Fei) was presented at the CVPR conference in 2009; the authors worked at Princeton University. The set is built on the WordNet hierarchy: for each synset, a group of synonyms for one concept, the authors aimed to collect 500–1,000 verified images on average. The first version had 12 subtrees, 5,247 synsets and 3.2 million images, and the labels were crowdsourced through Amazon Mechanical Turk. This scale made it possible to use the set as a benchmark for object recognition and image classification. From January 2017 to September 2018, Li was chief scientist of AI and machine learning at Google Cloud. In March 2019, the Stanford Institute for Human-Centered AI (Stanford HAI) launched, with her as founding director. Later, Li became one of the founders of World Labs, a spatial intelligence company that builds generative models of three-dimensional worlds.

Example

A packaging maker wants a camera on the line to reject dented boxes. Before training the model, the team collects and labels several thousand pictures of good and defective boxes: the same approach ImageNet is built on.

Related terms

Questions and answers

What is Fei-Fei Li known for? She created ImageNet, a set of millions of labeled photos used to test computer vision systems. Li and her co-authors presented the paper on the dataset at the CVPR conference in 2009.
Where does Fei-Fei Li work now? Li is a professor of computer science at Stanford and the founding director of the Stanford HAI institute. She is also among the founders of World Labs, a company that builds models of three-dimensional worlds.

Sources

Not sure whether your company needs this?

AI solutions map
Contact us