Computational Scientific Machine Learning Lab (CSML)

Computational Scientific Machine Learning Lab 🚀

Welcome to the official GitHub home of the CSML Lab at Rensselaer Polytechnic Institute (RPI). We reside at the intersection of Physics-Informed AI, Data-Driven Dynamical Systems, and Autonomous AI Scientist.

🌐 Lab Website | 📧 Contact | 📍 RPI MANE | 🚫 Lab Member Access

🧬 Our Core Research Pillars

Koopman Dynamics Koopman-Stable
Stable embeddings for dynamics and control.
- Phase I (2018) Long-Horizon + Jacobian Regularization
  - Long-time predictive modeling of nonlinear dynamical systems using neural networks
- Phase II (2019–2020) Stability + Time Delay Embedding
  - Physics-Informed Probabilistic Learning of Linear Embeddings of Non-linear Dynamics With Guaranteed Stability
  - On the Structure of Time-delay Embedding in Linear Models of Non-linear Dynamical Systems
- Phase III (2021) Modes Selection
  - Sparsity-promoting algorithms for the discovery of informative Koopman-invariant subspaces
- Phase IV (2024) Multi-Attractor + Open Source
  - On the lifting and reconstruction of nonlinear systems with multiple invariant sets
  - PyKoopman: A Python Package for Data-Driven Approximation of the Koopman Operator
- Phase V (2025) Noise-Robust Koopman and Control
  - Learning Noise-Robust Stable Koopman Operator for Control With Hankel DMD
Neural Representations Operator Learning
Mesh-agnostic neural fields and operator surrogates.
- Phase I (2019) CNN Surrogates
  - Prediction of Aerodynamic Flow Fields Using Convolutional Neural Networks
- Phase II (2020) PINNs
- Phase III (2023) Bridging INR with Operator Learning
  - Neural Implicit Flow: a mesh-agnostic dimensionality reduction paradigm of spatio-temporal data
- Phase IV (2024) Apply in Plasma
  - Grad–Shafranov equilibria via data-free physics informed neural networks
- Phase V (2025–2026) Apply in Nuclear
  - Surrogate modeling of heat transfer using Fourier Basis-DeepONet with uncertainty quantification
  - Characterization of DeepONet Performance for Neutron Transport Modeling
CFD and ROMs
Turbulence closures and high-rank reduced models.
- Phase I (2014–2015) High-Speed CFD
- Phase II (2016) Compressible Turbulence
  - The role of bulk viscosity on the decay of compressible, homogeneous, isotropic turbulence
- Phase III (2017) ML turbulence modeling
  - Characterizing and improving predictive accuracy in shock-turbulent boundary layer interactions using data-driven models
  - Augmentation of turbulence models using field inversion and machine learning
- Phase IV (2018) Data-Driven Closures
  - Data-driven Discovery of Closure Models
- Phase V (2025) A Near-Optimal Low-Rank Representation
  - Beyond the Kolmogorov Barrier: A Learnable Weighted Hybrid Autoencoder for Model Order Reduction
Agentic CFD Automation Foam-Agent
Agentic tools for OpenFOAM automation and data.
- Phase I (2025) Agents for CFD
  - Foam-Agent: Towards Automated Intelligent CFD Workflows
  - FoamGPT: Fine-Tuning Large Language Model for Agentic Automation of CFD Simulations with OpenFOAM
- Phase II (2025) Ecosystems + Infrastructure in the era of LLMs
  - Code2MCP: Transforming Code Repositories into MCP Services
  - UniFoil: A Universal Dataset of Airfoils in Transitional and Turbulent Regimes

🛠 Tech Stack & Tools

🌟 Featured Repositories

Project	Description	Status
PyKoopman	The premier Python library for Koopman operator learning.	⭐ Active
Foam-Agent	Agentic AI framework for automated OpenFOAM simulations.	🚀 New
NIF	Neural Implicit Flow for mesh-agnostic reduced-order modeling.	📄 Paper
CFDLLMBench	Benchmarking LLMs on Computational Fluid Dynamics tasks.	📊 Research