Example-Based Explainability in Machine Learning

Machine learning (ML) systems are increasingly utilized in high-stakes domains, where failures stemming from biases, data quality issues, or model mis-alignment can lead to severe consequences. To address these risks, explainable AI (XAI) techniques offer valuable insights into ML model decision-making. This chapter provides a comprehensive examination of XAI, with a particular focus on example-based methods, which aim to establish causal links between training data and model behavior, aiding in model debugging and data refinement. We present a taxonomy of XAI techniques, categorizing them into model-based, feature-based, and example-based approaches, and evaluate their strengths and weaknesses in terms of interpretability, adaptability, and robustness. The chapter delves into representative example-based methods, highlighting their key contributions, technical methodolo-gies, and inherent limitations. Furthermore, it critically reviews the state-of-the-art in example-based explainability, assessing how these methods address challenges such as model fidelity, user intent, and data quality. Finally, the chapter identifies open challenges and outlines future research directions, emphasizing the importance of developing user-centered, intent-aware explanations that align more closely with model behavior.