Greenfield and Brownfield Data Labeling – from Idea to Production with data-centric AI
"The unreasonable effectiveness of data" is a well-known phenomenon in data science, further extending concepts such as data-centric AI. This session will dive into the two differentiations of greenfield and brownfield labeling. The first is about prototyping and starting labeling from scratch, whereas the latter is about continuously enhancing existing data to improve a model’s performance. We will discuss scalable labeling with concepts such as weak supervision and active transfer learning, and further, consider techniques such as confident learning and neural search to address data management.