Representation Learning

When building an ML model, it is not clear which features need to be included in the model. Representation Learning (RL) allows the unbiased selection of features should be included in the model. When applied to chemical reactions, analysis of the features selected by RL can provide new insights into the reaction. C-CAS is developing new RL algorithms and applying them for the elucidation of reaction mechanisms, e.g. for Ni-catalyzed cross couplings

Publications

• Ma, C.; Yang, Q.; Gao, X.; Zhang, X. DEMO: Disentangled Molecular Graph Generation via an Invertible Flow Model. In: CIKM'22: Proceedings of the 31st ACM International Conference on Information & Knowledge Management. 2022, 1420-1429. https://doi.org/10.1145/3511808.3557217

• Zhichun Guo, Kehan Guo, Bozhao Nan, Yijun Tian, Roshni G. Iyer, Yihong Ma, Olaf Wiest, Xiangliang Zhang, Wei Wang, Chuxu Zhang, Nitesh V. Chawla. Graph-based Molecular Representation Learning. Intl. Joint Conf. Art. Int. 2023 , pp 6638-6646. https://doi.org/10.24963/ijcai.2023/744

• Tang P, Jiang M, Xia BN, Pitera JW, Welser J, Chawla NV. Multi-label patent categorization with non-local attention-based graph convolutional network. Proc. AAAI Conf. Art. Int.  2020 34, 9024-9031.   https://ojs.aaai.org/index.php/AAAI/article/view/6435

• Chen J, Guo K, Liu Z, Isayev O, Zhang X. Uncertainty-Aware Yield Prediction with Multimodal Molecular Features. In Proceedings of the AAAI Conference on Artificial Intelligence 2024 Mar 24 (Vol. 38, No. 8, pp. 8274-8282).