Hongkai Zheng
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Hongkai Zheng has a total of 18 co-authors. Here are the top co-authors by number of shared publications:
- Anima Anandkumar
Caltech
1 shared publication - Burigede Liu
Computing and Mathematical Science, California Institute of Technology
1 shared publication - Changhao Xu
Andrew and Peggy Cherng Department of Medical Engineering, Division of Engineering and Applied Science, California Institute of Technology, Pasadena, USA
1 shared publication - David Jin
Computing and Mathematical Science, California Institute of Technology
1 shared publication - Guanzhi Wang
Department of Computing + Mathematical Sciences, Division of Engineering and Applied Science, California Institute of Technology, Pasadena, CA, USA
1 shared publication - Haoxuan Chen
Computing and Mathematical Science, California Institute of Technology
1 shared publication
Latest Publications
- Open Access
- 31 July 2026
- Published by American Association for the Advancement of Science (AAAS) in Science Advances
- Vol. 12 (31)
Human-inspired time-series health evaluation with an adaptive multimodal electronic skin
Electronic skin powered with artificial intelligence could enable next-generation robotic and medical devices, yet integrating multimodal sensors and analyzing heterogeneous, multifrequency time series remain challenging. Most wearable machine learning architectures are time-invariant and trained for a specific task, limiting transfer across modalities and users. We present a multimodal electronic skin that captures diverse physiological signs with an adaptive learning framework that rapidly generalizes to unseen tasks with minimal labeled data. Our streamlined end-to-end framework uses a spectral variational autoencoder to denoise and compress multifrequency biosignals into a shared, unified second-wise latent space that preserves the spectral-temporal structure, followed by a transformer to capture temporal dependencies to support diverse downstream tasks with data-efficient learning. We demonstrate robust adaptation with 94.7% accuracy in activity recognition and 90.2% precision in fatigue assessment across various users and daily activities regardless of device and user variations, highlighting a scalable route to generalized physiological time-series analytics and human performance assessments.
- 31 July 2026
- Published by American Association for the Advancement of Science (AAAS) in Science Advances
- Vol. 12 (31)
Open AccessHuman-inspired time-series health evaluation with an adaptive multimodal electronic skin
Electronic skin powered with artificial intelligence could enable next-generation robotic and medical devices, yet integrating multimodal sensors and analyzing heterogeneous, multifrequency time series remain challenging. Most wearable machine learning architectures are time-invariant and trained for a specific task, limiting transfer across modalities and users. We present a multimodal electronic skin that captures diverse physiological signs with an adaptive learning framework that rapidly generalizes to unseen tasks with minimal labeled data. Our streamlined end-to-end framework uses a spectral variational autoencoder to denoise and compress multifrequency biosignals into a shared, unified second-wise latent space that preserves the spectral-temporal structure, followed by a transformer to capture temporal dependencies to support diverse downstream tasks with data-efficient learning. We demonstrate robust adaptation with 94.7% accuracy in activity recognition and 90.2% precision in fatigue assessment across various users and daily activities regardless of device and user variations, highlighting a scalable route to generalized physiological time-series analytics and human performance assessments.
- Top Cited PapersOpen Access
- 20 May 2024
- Published by Association for Computing Machinery (ACM) in ACM / IMS Journal of Data Science
- Vol. 1 (3) , 1-27
Physics-Informed Neural Operator for Learning Partial Differential Equations
In this article, we propose physics-informed neural operators (PINO) that combine training data and physics constraints to learn the solution operator of a given family of parametric Partial Differential Equations (PDE). PINO is the first hybrid approach incorporating data and PDE constraints at different resolutions to learn the operator. Specifically, in PINO, we combine coarse-resolution training data with PDE constraints imposed at a higher resolution. The resulting PINO model can accurately approximate the ground-truth solution operator for many popular PDE families and shows no degradation in accuracy even under zero-shot super-resolution, that is, being able to predict beyond the resolution of training data. PINO uses the Fourier neural operator (FNO) framework that is guaranteed to be a universal approximator for any continuous operator and discretization convergent in the limit of mesh refinement. By adding PDE constraints to FNO at a higher resolution, we obtain a high-fidelity reconstruction of the ground-truth operator. Moreover, PINO succeeds in settings where no training data is available and only PDE constraints are imposed, while previous approaches, such as the Physics-Informed Neural Network (PINN), fail due to optimization challenges, for example, in multi-scale dynamic systems such as Kolmogorov flows.
- 20 May 2024
- Published by Association for Computing Machinery (ACM) in ACM / IMS Journal of Data Science
- Vol. 1 (3) , 1-27
Top Cited PapersOpen AccessPhysics-Informed Neural Operator for Learning Partial Differential Equations
In this article, we propose physics-informed neural operators (PINO) that combine training data and physics constraints to learn the solution operator of a given family of parametric Partial Differential Equations (PDE). PINO is the first hybrid approach incorporating data and PDE constraints at different resolutions to learn the operator. Specifically, in PINO, we combine coarse-resolution training data with PDE constraints imposed at a higher resolution. The resulting PINO model can accurately approximate the ground-truth solution operator for many popular PDE families and shows no degradation in accuracy even under zero-shot super-resolution, that is, being able to predict beyond the resolution of training data. PINO uses the Fourier neural operator (FNO) framework that is guaranteed to be a universal approximator for any continuous operator and discretization convergent in the limit of mesh refinement. By adding PDE constraints to FNO at a higher resolution, we obtain a high-fidelity reconstruction of the ground-truth operator. Moreover, PINO succeeds in settings where no training data is available and only PDE constraints are imposed, while previous approaches, such as the Physics-Informed Neural Network (PINN), fail due to optimization challenges, for example, in multi-scale dynamic systems such as Kolmogorov flows.