Introduction to Data-Centric AI (dcai.csail.mit.edu)

A groundbreaking course on Data-Centric AI (DCAI) is set to launch as part of MIT's IAP 2024, addressing a crucial but often overlooked aspect of machine learning: improving datasets instead of solely focusing on model refinement. DCAI emphasizes that enhancing dataset quality can lead to better performance in real-world applications, countering the common "garbage in, garbage out" problem. This course will guide participants through systematic approaches to identify and rectify data issues, making it highly relevant for practitioners seeking practical solutions over theoretical knowledge. Running from January 16 to 26, 2024, the course features hands-on labs in Python to reinforce learning on topics such as label errors, class imbalance, and data curation for large language models (LLMs). While participants are encouraged to have a foundational understanding of machine learning and basic Python libraries, the course aims to be accessible to all interested in elevating their data handling skills, making it a significant step forward for the AI/ML community in adopting a more data-centered perspective.