Mori-Zwanzig Latent-Space Koopman Closure for Non-Linear Autoencoders
doi.org/10.1098/rspa.2024.0259 →
I am currently a PhD researcher in the Department of Aeronautics at Imperial College London, working at the intersection of machine learning, dynamical systems, and control theory. My research focuses on developing data-driven linear and bilinear models to enable efficient and accurate prediction and control of non-linear chaotic dynamical systems.
My work combines approaches from statistical physics, including Mori–Zwanzig and Koopman formalism, with modern machine learning techniques to create interpretable reduced-order models. I integrate these models into model predictive control (MPC) and reinforcement learning (RL) frameworks for adaptive, real-time control applications.
Please feel free to email me if you'd like to talk research or discuss collaboration :)
🔬 Research Interests
- Reduced-Order Modeling: Mori–Zwanzig latent-space Koopman closures for non-linear autoencoders, enabling interpretable reduced-order modeling of chaotic systems.
- Adaptive Control: adaptive model predictive control schemes that integrate interpretable ML with physics-based priors for real-time control of non-linear dynamics.
- Reinforcement Learning: ensemble and probabilistic model-based RL for control of chaotic systems, improving policy stability via value-function regularization.
- Physics-Informed ML: physics-constrained neural networks that enforce physically meaningful properties in latent spaces for improved generalization.
🎓 Education
-
PhD, Department of Aeronautics, Imperial College LondonThesis topic: computational modelling, numerical methods, and control of non-linear dynamical systems.
-
MSc, Aeronautics (Advanced Computational Methods), Imperial College LondonResult: Distinction.
-
B.Tech., Mechanical Engineering, Delhi Technological University (formerly DCE)CGPA: 8.78/10 (Distinction).
📚 Publications
Analysis of Generative Adversarial Networks for Data-Driven Inverse Airfoil Design
doi.org/10.1007/978-981-16-7618-5_22 →
Data-driven closure for Koopman control of non-linear systems
View presentation →
Model-based reinforcement learning for chaotic flow control
View presentation →
💼 Experience
-
Research Assistant, Imperial College LondonDeveloping Monte Carlo simulation methods for stochastic modelling of diffusion processes in complex geometries, with application to cardiac MRI; building computational workflows for ensemble particle simulations covering trajectory generation, statistical analysis, and quantitative model validation.
-
Graduate Teaching Assistant, Imperial College LondonConducted tutorials on computational methods, numerical simulation, and turbulence modelling.
-
MSc Research Project, Magri Lab, Imperial College LondonDeveloped physics-constrained non-linear autoencoders for dimensionality reduction of chaotic dynamical systems, incorporating conservation laws and symmetry constraints into the latent representation.
-
Research Intern, Monolith AI, LondonDesigned custom loss functions for neural networks to emphasize critical engineering design regions; built ML models for high-dimensional fluid dynamics problems (2D airfoil fields, 3D wind-turbine flows); developed chain-model architectures for scalar metric prediction.
-
Research Assistant, Delhi Technological UniversityDeveloped a data-driven inverse airfoil design framework using deep convolutional GANs; conducted a quantitative comparison of ML regressors for pressure field reconstruction from PIV data; designed a slat-airfoil shape optimization framework using evolutionary algorithms.