Research & Publications

Quantum-enhanced Inverse ECG Imaging Reconstruction (Aug 2026-present)

Methods: Hybrid Quantum-Classical Computing, Variational Quantum Linear Solver (VQLS), Block Encoding, Inverse Problems

Reference: Zhang, X., Lu, C., Yao, B.* (2026). Quantum-enhanced Inverse ECG Imaging Reconstruction (Preprint available online soon).

Designed a coherent VQLS that solves the encoded system on a quantum register, using a Möttönen ansatz for the trial state, a FABLE block encoding of the system matrix, and two projector read-outs for the cost function, and decoded the quantum solution to all heart nodes under a physical prior.
Showed that VQLS closely approximates the classical least-squares solution with very high fidelity in few optimization iterations, and that pre-encoding with a mesh-Laplacian regularizer improves reconstruction accuracy and stays robust as measurement noise increases.

P-K-GCN: Physics-augmented Koopman-enhanced Graph Convolutional Network for Deep Spatiotemporal Super-resolution (Aug 2025-present)

Methods: Physics Augmentation, Graph Neural Networks, Koopman Operator Theory, Deep Learning

Reference: Zhang, X., Wang, Z., Liu, F., Yao, B.* (2026). P-K-GCN: Physics-augmented Koopman-enhanced Graph Convolutional Network for Deep Spatiotemporal Super-resolution. IISE Transactions. (under review) [Paper]

Proposed P-K-GCN by integrating spline-based GCN, Koopman operator theory, and physics constraints for spatiotemporal super-resolution from coarse observational data.
Established generalization guarantees linking physics augmentation and Koopman regularization to reduced super-resolution error.
Outperformed standard data-driven baselines in reconstructing cardiac electrodynamics on complex 3D geometries.

Physics-augmented Multi-task Gaussian Process for Modeling Spatiotemporal Dynamics (Jan 2025-Jul 2026)

Methods: Physics Augmentation, Multi-task Learning, Gaussian Processes, Uncertainty Quantification

Reference: Zhang, X. & Yao, B.* (2026). Physics-augmented Multi-task Gaussian Process for Modeling Spatiotemporal Dynamics. IEEE Transactions on Automation Science and Engineering, 23, 13316-13329. [Paper]

Developed geometry-aware multi-task Gaussian Processes (G-MGP) to jointly model spatiotemporal structure and cross-task dependencies on 3D manifolds.
Extended G-MGP to physics-augmented G-MGP (P-G-MGP) to incorporate underlying physical laws, thereby constraining predictions to be consistent with governing dynamical principles.
Provided closed-form estimates of posterior variance, enabling calibrated uncertainty quantification for downstream decision-making.

Geometry-aware Active Learning for Spatiotemporal Systems (Aug 2024- Feb 2026)

Methods: Manifold Learning, Gaussian Processes, Active Learning

Reference: Zhang, X. & Yao, B.* (2026). Geometry-aware Active Learning of Spatiotemporal Dynamic Systems. IISE Transactions, 1–22. [Paper]

Proposed Geometry-aware Spatiotemporal Gaussian Process (G-ST-GP) for forecasting high-dimensional dynamics on non-Euclidean 3D domains.
Developed an adaptive active learning (A-AL) strategy to strategically identify informative spatial locations for training data collection and further maximize the prediction accuracy.
Implemented scalable Kronecker/tensor inference to reduce training and prediction cost on large spatiotemporal datasets.

Methodological Evaluation of Assessment Models for Ocean Energy Systems (Sep 2019-Jun 2021)

Methods: Dynamic Life Cycle Assessment (LCA), Renewable Energy Assessment, ISO Standardization

Reference: Zhang, X., Zhang, L., Yuan, Y., & Zhai, Q.* (2020). Life Cycle Assessment on Wave and Tidal Energy Systems: A Review of Current Methodological Practice. International Journal of Environmental Research and Public Health, 17(5), 1604. [Paper]

Audited 18 renewable energy assessment models for ISO compliance and decision-grade analytics for industry stakeholders.
Identified limitations of static models in long-horizon forecasting and their impact on investor risk assessment and capital allocation.
Advocated Dynamic LCA frameworks to improve 20+ year predictive accuracy for sustainability and investment decisions.

More works are in progress. Please stay tuned !

Page updated

Google Sites

Report abuse