platform.ai allows domain experts to produce high-quality labels in minutes in a visual, interactive fashion. Jeremy Howard (platform’s Chief Scientist) initially demonstrated this concept in a TEDx talk, where he built a model in 15 minutes that can classify over a million unlabeled points to 99% accuracy (something that previously could take days or weeks). Our motivation was to build an application that non-technical people can use to train good deep learning models.
An extensive amount of research work has gone into platform including leveraging human perception, active learning, transfer from pre-trained nets, and noise-resilient training. We want to use the labeler's time in the most productive way and have the model learn from every aspect of the human interaction.
We believe this approach is broadly applicable to many industry use cases including autonomous driving, facial recognition, defense, product recommendations, medical imaging diagnostics, industrial quality control, etc. and that platform.ai can significantly accelerate the development of models in all these areas by providing pre-trained networks and allowing collaboration among domain experts.