Emerging Development and Challenges in Data-Centric AI
Chaitali Shewale · 2024
The chapter digs into the tenets and elements of data-centric AI (DCAI), with a particular emphasis on methodically modifying data to improve model performance. It covers data gathering, preprocessing, model training, and data augmentation, emphasizing the significance of efficiently managing and exploiting data throughout the AI development lifecycle. To ensure the appropriateness of the data used, the evaluation of data quality and applicability for AI applications is prioritized. The chapter also emphasizes the importance of data management in data-centric AI to maintain data quality and relevance over time. It introduces model-centric AI and contrasts it with the data-centric approach, which emphasizes systematic data enhancement for AI systems, by emphasizing its focus on model design. Data corruption, imbalance, privacy issues, biases, and other difficulties in data-centric AI are discussed, highlighting the need for efficient approaches to handle these difficulties and improve AI systems. The chapter illustrates the shift toward efficient data handling and AI system optimization by comparing the differences between model-centric and data-centric AI.