My research sits at the intersection of scientific computing and machine learning, with a focus on building fast, scalable, and differentiable solvers for partial differential equations (PDEs).
- Differentiable solvers: I created TensorMesh (finite elements) and torch-sla (sparse linear algebra), bringing classical numerical methods into PyTorch.
- Scalable neural solver backbones: neural operators for complex geometries, such as GAOT and RIGNO, accelerated with fused GPU kernels such as FlashSlice.
- Physics-informed operator learning: training neural operators from the governing equations alone; multigrid preconditioning lets this label-free training match supervised accuracy.
- PDE foundation models: large-scale pretraining across diverse physical systems
- Preconditioned Physics-Informed Neural Operator Training: S. Wen*, S. Mishra, M. Zeinhofer* · arXiv 2026 · paper · code
- Does Transolver really need a Transformer?: S. Wen, S. Mishra · arXiv 2026 · paper · code
- Learning, Solving and Optimizing PDEs with TensorGalerkin: an efficient high-performance Galerkin assembly algorithm: S. Wen*, M. Chi*, T. Yu, B. Moseley, M. Y. Michelis, P. Ren, H. Sun, S. Mishra · ICML 2026 · paper · code · project
- torch-sla: Differentiable Sparse Linear Algebra with Adjoint Solvers and Sparse Tensor Parallelism for PyTorch: M. Chi*, S. Wen* · ICML 2026 Workshop (AI4Physics) · paper · code · project
- Geometry Aware Operator Transformer as an Efficient and Accurate Neural Surrogate for PDEs on Arbitrary Domains: S. Wen, A. Kumbhat, L. Lingsch, S. Mousavi, Y. Zhao, P. Chandrashekar, S. Mishra · NeurIPS 2025 · paper · code · project
Open to collaborations on scientific ML, neural operators and differentiable simulation.