<ul><li>Recourse generators provide actionable insights, often through feature-based counterfactual explanations (CFEs).</li><li>Introducing three novel recourse types grounded in real-world actions: high-level continuous (hl-continuous), high-level discrete (hl-discrete), and high-level ID (hl-id) CFEs.</li><li>Proposing data-driven CFE generation approaches that quickly provide optimal CFEs for new agents.</li><li>Empirical evaluation shows the effectiveness of the proposed forms of recourse over low-level CFEs.</li></ul>

Learning Actionable Counterfactual Explanations in Large State Spaces

Discover more