Statistical Foundations for Trustworthy AI

Integrating statistical reasoning, principles, and techniques with AI is crucial for advancing health and public health applications. While AI offers powerful tools for analyzing large and complex datasets, statistical methodologies provide the rigor and structure needed to ensure the validity, reliability, and generalizability of these models. By applying statistical reasoning, we can move beyond simple predictions to understand the underlying relationships in the data, drawing causal inferences that inform actionable interventions.

This integration is particularly important for health applications, where decisions impact patient outcomes and public policy. TRAIL dedicates its research in this direction to ensure AI is trustworthy, fair, and applicable to diverse populations, making AI solutions more robust, transparent, and effective in improving health outcomes.

View our featured work and publications

1. Liu Q., Wang, Z., Li, X., Ji, X., Zhang, L., Liu, L., and Liu Z.  (2024). DNA-SE: Towards deep neural-nets assisted semiparametric estimation. Proceedings of the 41st International Conference on Machine Learning (ICML), 235 :32041-32061. 
2. Xu, S., Liu, L. and Liu, Z.  (2022) DeepMed: Semiparametric causal mediation analysis with debiased deep learning. The Thirty-sixth Conference on Neural Information Processing Systems (NeurIPS), 35, pp. 28238-28251.