Concept Selection

Concept selection in machine learning focuses on improving the interpretability and efficiency of models by identifying and utilizing a subset of key concepts for prediction, rather than relying on complex, "black-box" approaches. Current research emphasizes automated concept discovery methods, often leveraging vision-language models and employing techniques like score matching and concept embedding approximations to select the most informative concepts. This work aims to enhance model transparency, reduce computational costs, and allow for stakeholder customization of the concepts used in decision-making, ultimately leading to more trustworthy and understandable AI systems.