Explore a one-hour conference talk by Shantanu Ghosh from Stanford University on developing concept-based interpretable models for efficient transfer learning in healthcare AI. Dive into the challenges of building generalizable AI models for medical imaging and learn about a novel approach that combines blackbox neural networks with interpretable components. Discover how this method iteratively carves out concept-based models using First Order Logic, potentially improving generalizability and reducing the need for extensive labeled data in target domains. Gain insights into the speaker's research on blurring the distinction between post-hoc explanations and interpretable model construction, and understand the implications for enhancing AI model flexibility, explainability, and transfer efficiency in medical applications.