API World + AI DevWorld

AI technology can deliver huge value, but the perception is this value is primarily achievable only by those with huge training datasets and the ability to hire a team of highly skilled AI scientists and engineers. In this talk I will describe the impact of empowering developers by 1. training useful models with smaller datasets and 2. making computer vision model training a service as well as open source. I will share the lessons learned from building a public cloud API in which users around the world trained tens of thousands of different vision models, in terms of both the development process and the user experience. I will also describe the impact of a new open source toolkit called "Learning with Less Labels" which brings multiple assets directly from research papers into developers' hands to enable effective learning from hundreds of images, instead of millions.The goal is making computer vision work for every developer. Where will the path lead when we are no longer in an ImageNet-1K monoculture?