01Scientific machine learning / AI for Science
ChenhaoSi
司辰昊
PhD Candidate · Year 5
School of Data Science · CUHK-Shenzhen
I develop numerical learning methods for PDE-governed systems—making physics-informed neural networks more accurate, scalable, and easier to train.
Research coordinates
- 01Scientific ML
- 02Physics-informed learning
- 03PDE solvers
2026Selected work / 01
A single Cauchy layer for complex, high-dimensional PDEs.
compleX-PINN introduces a learnable activation inspired by the Cauchy integral formula, delivering high accuracy with a compact architecture.
Scholar metricsLast successful sync: 2026-09-30 13:45 (Asia/Shanghai, UTC+8)
- citations
- 61
- h-index
- 3
- listed works
- 7
03Selected publications / 07
Architectures, optimization, and generative scientific models.
Selected articles, workshop work, and preprints. Scholar metrics and citation counts are checked automatically each day; new papers are added here after review. An unavailable citation count is shown as —; see Google Scholar for the complete record.
- 012026JOURNALConvolution-weighting method for the physics-informed neural network: A primal-dual optimization perspectiveJournal of Computational Physics 555, 114773
- 022026JOURNALComplex physics-informed neural networkJournal of Computational Physics 553, 114713CODE ↗
- 032026PREPRINTFrom Non-Convex Self-Concordant Regularization to Scalable Quasi-Newton Training of PINNsarXiv:2608.04206
- 042026PREPRINTLightweight Geometric Adaptation for Training Physics-Informed Neural NetworksarXiv:2604.15392
- 052025JOURNALInitialization-enhanced physics-informed neural network with domain decomposition (IDPINN)Journal of Computational Physics 530, 113914
- 062025WORKSHOPAPOD: Adaptive PDE-observation diffusion for physics-constrained samplingICML Workshop on Assessing World Models
- 072025PREPRINTAutoBalance: An automatic balancing framework for training physics-informed neural networksarXiv:2510.06684
04Position / trajectory
Building practical learning systems for hard scientific problems.
My work sits between numerical analysis and machine learning. I study how structure from differential equations can guide models—and how optimization can make those models reliable on difficult, high-dimensional systems.
PhD in Data Science
School of Data Science · CUHK-Shenzhen
Working focus
- PINN architecture
- Adaptive optimization
- High-dimensional PDEs
- Physics-constrained generation
05Open channels
Ideas, questions, collaborations?
I am always interested in thoughtful conversations around scientific machine learning, numerical methods, and AI for Science.