Learning to Optimize (L2O)
Accelerating the solution of complex, constrained optimization problems through solver-free, self-supervised learning frameworks.
Learning to Optimize for Mixed-Integer Non-linear Programming with Feasibility Guarantees
L2O framework addressing parametric MINLP by integrating integer correction layers for adaptive rounding with projections to ensure feasibility
Learning to Solve Constrained Bilevel Control Co-Design Problems
Self-supervised L2O framework solving constrained bilevel control co-design problems trough control with enforced coupling constraints
Learning to Control (L2C)
Synthesizing constrained neural control policies for nonlinear and infinite-dimensional systems via differentiable predictive control and offline self-supervised learning.
Learning to Solve Parametric Mixed-Integer Optimal Control Problems via DPC
Achieving near-optimal control performance via DPC in input- and state-constrained parametric mixed-integer optimal control problems.
Zero-Shot Function Encoder-Based Differentiable Predictive Control
Integrating Function Encoder-based Neural ODEs with DPC to enable zero-shot control of nonlinear systems
Learning to Control PDEs with Differentiable Predictive Control and Time-Integrated Neural Operators
End-to-end framework integrating Time-Integrated DeepONets with DPC to enable the offline, self-supervised optimization of control policies for PDEs
Model Predictive Control (MPC)
Advancing the computational efficiency, real-time execution, and predictive accuracy of linear and nonlinear Model Predictive Control systems through novel solver architectures and data-driven dynamics learning.
A Time-Certified Predictor-Corrector IPM Algorithm for Box-QP
πMPC: A Parallel-in-horizon and Construction-free NMPC Solver
The MPC algorithm delivers a construction-free, parallel-in-horizon nonlinear model predictive control solver by combining a novel variable-splitting scheme with a velocity-based system representation to enable highly efficient execution directly on system matrices.
Least-Squares Multi-Step Koopman Operator Learning for Model Predictive Control
Learning long-horizon dynamics thorugh convex multi-step Koopman framework while completely bypassing the compounding prediction errors that plague traditional single-step models.
Differentiable Programming
Integrating differentiability into engineering modeling frameworks to enable gradient-based learning, verification, and design optimization.
∂CBDs: Differentiable Causal Block Diagrams
Integrating causal block diagrams, differentiability for learning and contract-based verification frameworks.
Energy systems
Enhancing the operation and scheduling of energy storage and grid systems through decision-focused learning and advanced optimization.
Accelerating Underground Pumped Hydro Energy Storage Scheduling with Decision-Focused Learning
This paper presents a decision-focused learning framework for underground pumped hydro energy storage scheduling that uses neural networks to guide recursive linearization.
Scientific Machine Learning (SciML)
Bridging physics-based modeling with machine learning to ensure safety, stability, and physical consistency in data-driven representations of dynamical systems.
Safe Physics-informed Machine Learning for Dynamics and Control
Tutorial paper provides a comprehensive overview of safe physics-informed machine learning for dynamics and control.
Nonlinear System Identification
Advancing data-driven modeling of complex dynamics.