Learn about gradient-based input attribution methods in machine learning through a comprehensive lecture that progresses from foundational concepts to advanced applications. Begin with a review of free-text explanations before exploring the critical concept of faithfulness in attribution methods. Dive deep into the motivation behind gradient-based attribution and highlighting techniques, followed by detailed computational methods illustrated through practical examples. Examine the limitations of these approaches and discover various extensions that address these constraints. The lecture includes visual demonstrations and concludes with a thorough recap of key concepts, making complex attribution techniques accessible for both beginners and experienced practitioners.