What this book covers
Chapter 1, Exploring Data-Centric Machine Learning, contains a comprehensive definition of data-centric machine learning and draws contrasts with its counterpart, model-centricity. We use practical examples to compare empirical performance and illustrate key differences between these two methodologies.
Chapter 2, From Model-Centric to Data-Centric – ML’s Evolution, takes you on a journey through the evolution of AI and ML toward a model-centric approach, highlighting the untapped potential in improving data quality over model tuning. We also debunk the “big data” myth, showing how shifting to “good data” can democratize ML solutions. Get ready for a fresh perspective on the power of data in ML.
Chapter 3, Principles of Data-Centric ML, sets the stage for your journey into the heart of data-centric ML by outlining the four key principles of data-centric ML. These principles offer crucial context – the why – before we delve into the specific methods and approaches linked to each principle – the what – in the ensuing chapters.
Chapter 4, Data Labeling Is a Collaborative Process, explores the pivotal role of subject-matter expertise, trained labelers, and clear instructions in ML development. In this chapter, you will learn about the human-centric nature of data labeling and acquire strategies to enhance it to reduce bias, increase consistency, and build richer datasets.
Chapter 5, Techniques for Data Cleaning, explores the six crucial aspects of data quality and showcases various techniques for cleaning data, a vital process for enhancing data quality by rectifying errors. We illustrate why questioning and systematically improving data quality is crucial for reliable machine learning systems, all while teaching you essential data cleaning skills.
Chapter 6, Techniques for Programmatic Labeling in Machine Learning, focuses on programmatic labeling techniques for boosting data quality and signal strength. We go through the pros and cons of programmatic labeling and provide practical examples of how to execute and validate these techniques.
Chapter 7, Using Synthetic Data in Data-Centric Machine Learning, introduces synthetic data as an efficient and cost-effective method for overcoming the limitations of traditional data collection and labeling. In this chapter, you will learn what synthetic data is, how it’s used to improve models, the techniques to generate it, and its risks and challenges.
Chapter 8, Techniques for Identifying and Removing Bias, focuses on the problem of bias in the way we collect data, apply data and models to a problem, and the inherent human bias captured in many datasets. We will go through data-centric techniques for identifying and correcting biases in an ethical manner.
Chapter 9, Dealing with Edge Cases and Rare Events in Machine Learning, explains the process of detecting rare events in ML. We explore various methods and techniques, discuss the importance of evaluation metrics, and illustrate the wide-ranging impacts of identifying rare events.
Chapter 10, Kick-Starting Your Journey in Data-Centric Machine Learning, sheds light on the technical and non-technical challenges you might face during model development and deployment. This final chapter shows you how a data-centric approach can help you overcome these challenges, opening up big opportunities for growth and wider use of machine learning in your organization.
