data-centric-ai

Why we included this project

Most teams that ship machine learning eventually find the data is the bottleneck, not the model. This repository, maintained by Stanford's Hazy Research group, collects the writing on that problem in one place, organized into sections on weak supervision, data augmentation, data cleaning, robustness, and fine-grained evaluation, each pointing to the papers, talks, and tools that anchor the topic. It is an admittedly opinionated index rather than software you install, so it works best as the start of a reading list for anyone trying to improve data quality in their own pipelines. The maintainers also take contributions from readers, which keeps the list tied to the field as it moves.