I am an Assistant Professor of Aerospace Engineering at Rensselaer Polytechnic
Institute, where I lead the Computational Scientific Machine Learning (CSML) Lab. I work on the
frontiers of AI for physical systems by developing transformative numerical and data-driven methods for
predictive modeling, inverse problems, and control of complex large-scale dynamical systems, ultimately
accelerating scientific discovery across engineering and applied sciences.
I earned my Ph.D. in Aerospace Engineering and Scientific Computing from the University of Michigan (2021), and completed postdoctoral research at the
AI Institute in Dynamic Systems at the University of Washington. I am
affiliated with the Rensselaer-IBM AI Research Collaboration. Our lab has
access to numerous high-end HPC clusters (A100/A6000/GB200).
- Agentic AI for Computational Science (7 papers) Scientific tools, adaptive workflows, and evaluation for agents that carry out computational tasks.
- Scientific Machine Learning (13 papers) Methods and datasets for physical modeling, inverse problems, and numerical computation.
- Koopman Operators & Data-Driven Control (6 papers) Koopman models, time-delay embeddings, and stable dynamics for prediction and control.
- Representation Learning for PDE Systems (2 papers) Compact, mesh-agnostic representations and nonlinear model reduction for PDE solution fields.
- Compressible Turbulence / Flow Physics / Rarefied Gas Dynamics (5 papers) Compressible turbulence, rarefied gas dynamics, and industrial flow modeling.
- AI for Science in General (2 papers) Biomedical language models and efficient medical image segmentation.
- 10/2026 Aoming Liang joined the CSML Lab as a postdoctoral researcher. Welcome!
- 09/2026 Ph.D. student Rabi Pathak received the APS DFD Enabling Grant to attend the 79th APS DFD Annual Meeting
- 09/2026 Published a co-authored paper on post-training quantization for medical image segmentation foundation models (GPTQ-MedFM) in Medical Image Analysis, enabling clinical edge deployment of large segmentation models without accuracy loss [Paper]
- 08/2026 Ph.D. student Ling Yue published a first-author survey in EMNLP 2026 (Findings), establishing a unified Agentic Computation Graph (ACG) framework to systematically classify static pipelines and dynamic runtime workflow optimization for LLM agents [Paper]
- 08/2026 Ph.D. student Ling Yue published a paper as first author in Computer Methods in Applied Mechanics and Engineering (earlier at NeurIPS 2025 Workshop ML4PS) introducing Foam-Agent, an end-to-end multi-agent framework that automates OpenFOAM simulation setup and runtime error recovery with self-correcting feedback [Paper] [Code]
- 07/2026 Our collaboration with Shimin Di's group on self-evolving agents (FlowEvo) has been accepted to COLM 2026, proposing a framework for agents to autonomously self-evolve through the co-evolution of high-level workflows and executable code skills [Paper] [Code]
- 06/2026 Ph.D. student Ling Yue published a paper as first author in Health Data Science presenting BioMamba, domain-adapted state-space models that overcome the quadratic bottleneck of Transformers on long biomedical texts with linear-time inference [Paper]
- 06/2026 Ph.D. student Sai Bhargav Pochinapeddi received the Victor Peng '81 Summer Research Project Award
- 06/2026 Ph.D. students Ling Yue and Nithin Somasekharan are featured in the RPI Graduate Student Student Spotlight for their research in interdisciplinary AI and scientific machine learning
- 05/2026 Our collaboration with Min-Ling Zhang and Shimin Di's group on Code2MCP (earlier at NeurIPS 2025 Workshop SEA) has been published in KDD 2026, establishing an autonomous multi-agent pipeline that converts arbitrary GitHub software repositories into standardized Model Context Protocol (MCP) tool servers [Paper] [Code]
- 05/2026 Published a paper on neural-operator surrogate modeling for neutron transport in Nuclear Science and Engineering, replacing computationally intensive transport sweep loops with single-pass neural operators to achieve over 1,000× speedup while preserving eigenvalue accuracy within ~100 pcm [Paper]
- 04/2026 Ph.D. student Ling Yue published a perspective paper as first author in Frontiers in Artificial Intelligence, laying out a roadmap for scaling autonomous scientific discovery via hierarchical, protocol-native (MCP) multi-agent ecosystems [Paper]
- 03/2026 Received NVIDIA Academic Grant Program Award
- 03/2026 Published a co-authored paper in Journal of Computational Physics introducing MD-PNOP, which recasts governing equations into neural operators to achieve accurate parametric extrapolation from minimal data and accelerate iterative PDE solvers [Paper]
- 02/2026 Ph.D. student Nithin Somasekharan published a first-author paper in Journal of Data-centric Machine Learning Research releasing CFDLLMBench, the first comprehensive benchmark evaluating LLMs across domain knowledge, code generation, and simulation setup in fluid dynamics [PDF] [Code]
- 01/2026 Yangyuan Li joined the CSML Lab as a new Ph.D. student. Welcome!
- 01/2026 Gave a talk on "Deep Koopman sensing" at Joint Mathematics Meetings in Washington, D.C. [Abstract]
- 01/2026 Published a paper in Nuclear Science and Engineering characterizing DeepONet for neutron transport, enabling zero-retraining, real-time angular flux prediction under arbitrary source distributions with up to 900× speedup for reactor digital twins [Paper]
- 12/2025 Ph.D. student Ling Yue presented a first-author paper on multi-agent systems for CFD at NeurIPS 2025 Workshop ML4PS, exploring agent collaboration for automated OpenFOAM case construction and execution [Paper]
- 12/2025 Ph.D. student Ling Yue presented a first-author paper introducing FoamGPT at NeurIPS 2025 Workshop ML4PS, domain-tuning LLMs on specialized CFD instructions to eliminate syntactic hallucinations in OpenFOAM dictionary generation [Paper]
- 12/2025 Our collaboration with Sicheng He's group on UniFoil has been published at NeurIPS 2025, providing a benchmark dataset of 500k+ high-fidelity RANS simulations across laminar-turbulent transition and subsonic-transonic aerodynamic regimes [Paper] [Code]
- 09/2025 Ph.D. student Shahriar Akbar Sakib published a paper as first author in IEEE Transactions on Control Systems Technology, developing a stable parameterization with Hankel observables to resolve the noise sensitivity and spectral instability of data-driven Koopman operators for control [Paper]
- 08/2025 Ph.D. student Nithin Somasekharan published a paper as first author in Proceedings of The Royal Society A, introducing a learnable weighted hybrid autoencoder that overcomes the Kolmogorov barrier of linear reduction while guaranteeing SVD-like spectral convergence [Paper].
- 08/2025 Published a paper in Progress in Nuclear Energy, combining Fourier-basis DeepONet with Bayesian uncertainty quantification to accurately capture oscillatory heat transfer under coolant flow fluctuations in nuclear systems [Paper].
- 06/2025 Received Google Research Scholar Award [Link]
- 05/2025 Presented a work on "Toward Intelligent CFD Workflows in the Era of Large Language Models" at Algorithms For Multiphysics Models In The Post-Moore's Law Era in Los Alamos [Abstract]
- 05/2025 Presented a work on "Deep Koopman Sensing" at 1st International Symposium on AI and Fluid Mechanics [Abstract]
- 04/2025 Ph.D. student Nithin presented a work on "Deep Koopman Sensing" at 2nd ERCOFTAC Workshop on Machine Learning for Fluid Dynamics [Abstract]
- 04/2025 Ph.D. student Shahriar Akbar Sakib released a preprint on noise-robust Koopman operator learning, leveraging Hankel-embedded observables and progressive rollout loss to enforce spectral stability under measurement noise [Paper]
- 11/2024 Ph.D. student Weichao presented a work on meshless surrogate models with stability guarantees at APS-DFD 2024 [Abstract].
- 11/2024 Co-advised Ph.D. student Isaac presented a work on data-driven surrogate models that accelerate fluid solver at APS-DFD 2024 [Abstract].
- 11/2024 Ph.D. student Nithin Somasekharan presented work on convergence-guaranteed autoencoders at APS-DFD 2024, pairing deep networks with SVD weighting to guarantee rank convergence for chaotic multiscale flows [Abstract] [Paper]
- 11/2024 Ph.D. student Weichao presented a work on meshless surrogate models with stability guarantees at MORE 2024 [Abstract].
- 11/2024 Ph.D. student Shahriar presented a work on learning Koopman operator from data at SIAM New York-New Jersey-Pennsylvania Section Conference [Abstract].
- 10/2024 Presented a work on meshless surrogate models with stability guarantees at AMS Fall Eastern Sectional Meeting [Abstract].
- 09/2024 Published a co-authored paper in Journal of Geophysical Research: Machine Learning and Computation on distributed deep compression for seafloor acoustic sensing, utilizing recurrent decoders for high-ratio lossy compression and ocean-seismic wavefield separation [Paper]
- 07/2024 Presented a work on meshless surrogate models with stability guarantees at WCCM 2024
- 07/2024 Presented a conference paper at AIAA Aviation 2024 on composite Koopman analysis of hypersonic nonequilibrium flows, uncovering dynamical linkages between shock-boundary layer interactions and wall heat flux fluctuations [Paper]
- 07/2024 Published a co-authored paper at AIAA Aviation 2024 on learning incomplete LU (ILU) preconditioners via graph neural networks, accelerating iterative linear solver convergence for advection-dominated problems [Paper]
- 03/2024 Published a paper in Nonlinear Dynamics clarifying linear reconstruction mechanisms for systems with multiple invariant sets, demonstrating that discrete symmetries enable highly data-efficient global Koopman operator learning [Paper]
- 03/2024 Published a co-authored paper in Physics of Plasmas ([Editor's Pick]), introducing data-free and geometry-parameterized PINNs to solve the Grad–Shafranov equation for rapid tokamak plasma equilibrium calculations without meshing [Paper]
- 12/2023 Published a paper in Journal of Open Source Software introducing PyKoopman, a unified, scikit-learn-compatible Python package for data-driven approximation of the Koopman operator with modular observable dictionaries [Paper] [Code]
- 11/2023 Ph.D. student Nithin presented a work on differentiable finite-element solver using Nvidia Warp at APS-DFD 2023 [Abstract]
- 01/2023 Published a paper in Journal of Machine Learning Research introducing Neural Implicit Flow (NIF), a mesh-agnostic dimensionality reduction paradigm that breaks free from fixed grid topologies by decoupling spatial coordinates from spatio-temporal dynamics [Paper]
- Technical Committee: AIAA-FD, ICCS-TAMCS, IEEE Symposium
Series On Computational Intelligence 2024-present
- Panelist: NSF 2024
- Member: AIAA, AMS, APS, U.S. Association for Computational
Mechanics 2016-present
