Uncertainty Quantification (UQ)
Uncertainty quantification is the process by which a model evaluates how "sure" it is of its prediction. This is key at the deployment of the model since understanding where the model may have lower accuracy is important for selection of candidates for further evaluation as well as understanding what type of dataset augmentation may be useful.
Publications
Guo, K.; Liu, Z.; Guo, Z.; Nan, B.; Isayev, O.; Chawla, N.V.; Wiest, O.; Zhang, X. Proto-Yield: An Uncertainty-Aware Prototype Network for Yield Prediction in Real-world Chemical Reactions. 34th ACM Intl. Conf. Inf. Knowl. Man. (CIKM’25) 2025 accepted.
Dong, Kaiwen, Zhichun Guo, and Nitesh Chawla. Pure message passing can estimate common neighbor for link prediction. Advances in Neural Information Processing Systems 37 (2025): 73000-73035.
Chen, J., Guo, K., Liu, Z., Isayev, O. and Zhang, X., 2024, March. Uncertainty-Aware Yield Prediction with Multimodal Molecular Features. Proc. AAAI Conf. AI 2024 38, 8274-8282. https://doi.org/10.1609/aaai.v38i8.28668