I am a senior data scientist in the SciML group. I have been working on both novel theoretical/methodological developments and real-world problem solving with the state-of-the-art machine learning technologies. My recent research is focused on two directions: 1) physics-informed machine learning, aimed at improving the accuracy and generalisation of neural networks by building physics (such as differential equations) into them, and 2) enhanced self-attention mechanisms for leveraging transformer-based models. I was granted a PhD degree in Civil Engineering by Tsinghua University and then another PhD degree in Geophysics by the University of Oxford. Before joining SciML as a senior data scientist, I worked as a postdoc researcher at Yale University and the University of Oxford.
Latest first-authored publications:
[1] Leng, K., Shankar, M., & Thiyagalingam, J. (2024). Zero Coordinate Shift: Whetted Automatic Differentiation for Physics-informed Operator Learning. Journal of Computational Physics, 112904.
[2] Leng, K., Atwood, R. C., Kockelmann, W., Chekrygina, D., & Thiyagalingam, J. (2024). Unsupervised Image Segmentation with Dense Feature Learning and Sparse Clustering. International Conference on Computer Vision Theory and Applications, 2024.
[3] Leng, K., & Thiyagalingam, J. (2023). Padding-free Convolution based on Preservation of Differential Characteristics of Kernels. IEEE International Conference on Machine Learning and Applications, 2023.
[4] Leng, K., & Thiyagalingam, J. (2022). On the Compatibility between Neural Networks and Partial Differential Equations for Physics-informed Learning. arXiv preprint arXiv:2212.00270.
[5] Leng, K., King, S., Snow, T., Rogers, S., Markvardsen, A., Maheswaran, S., & Thiyagalingam, J. (2022). Parameter inversion of a polydisperse system in small-angle scattering. Journal of Applied Crystallography, 55(4), 966-977.
