Computational Autonomy
Our work
We develop approaches to automate reasoning and decision making in the presence of uncertainty. Our approaches focus on foundational algorithm development — we seek to create scalable, general, and adaptive algorithms that can intelligently work across many different application regimes. These goals require using and developing new tools in the areas of uncertainty quantification, statistical inference, and numerical analysis. We then apply these tools across areas ranging from data analysis and machine learning, to stochastic optimal control and optimization.
News
2023
- July 23-27 Doruk, Liliang and Thomas presented their work at the 17th US National Congress on Computational Mechanics in Albuquerque, New Mexico. Doruk's presentation was about "Solving Inverse Problems in High-Dimensional Data: An Approach for Optimal Sensor Placement with Incremental Tensor-Train Decomposition." Liliang's presentation was "A Novel Online Variational Inference Algorithm for State-Parameter Estimation of Hidden Linear Models." Thomas' presentation was on multi-output approximate control variates.
- May Nick has defended his PhD Thesis "Generalization of system identification objective functions through stochastic hidden Markov models for regularization, smoothness, and uncertainty quantification." Congratulations Nick!
- May Alex won an NSF CAREER Award: "CAREER: New Foundations for Multi-Fidelity Prediction, Estimation, and Learning Under Uncertainty in Dynamical Systems"
- May 24-26 Alex gave a talk on "Algorithms and approaches for multifidelity uncertainty quantification" at the ASA/IMS Spring Research Conference 2023 at Banff Centre, Alberta, Canada.
- April Josh won a NASA Space Technology Graduate Research Opportunities (NSTGRO) fellowship. Congratulations Josh!
- April 6-7 Alex gave an invited talk on "Methods and approaches to benchmark data-driven modeling in sparse and noisy data-regimes" at the USACM Workshop on Establishing Benchmarks for Data Driven Modeling of Physical Systems at the University of Southern California.
- March Alex gave an invited talk on "Opportunities and challenges for uncertainty quantification in the analysis, design, and control of complex engineering systems" at NASA Langley Research Center for the Uncertainty Quantification Seminar Series.
- March 24 Doruk won second prize at the Michigan Institute for Computational Discovery & Engineering (MICDE) Annual Symposium. Congratulations Doruk!
- Feburary 26 - March 3 Doruk, Thomas, and Alex presented their work at SIAM CSE in Amsterdam. Doruk presented the poster "An incremental tensor-train decomposition for high-dimensional data streams", the corresponding preprint is available here. Thomas presented his work on Multi-output Approximate Control Variates. Alex presented on "Graph Construction in Multi-Fidelity Networked Surrogates."
- January 23-27 Carleen and Josh presented their papers at AIAA Scitech. Carleen's paper was on "Online parameter estimation with trajectory optimization for dynamic soaring," and can be found here. Josh's paper was "Optimal experimental design to learn reduced-fidelity models for porous electrosprays", a collaboration with PEPL that can be found here. Alex also has had two other papers collaborating with Sandia National Laboratories and NASA on multifidelity uncertainty quantification approaches.
2022
- December 6–9 Nick's paper "Bayesian Identification of Nonseparable Hamiltonian Systems Using Stochastic Dynamic Models" was published in the proceedings of the Conference on Decision and Control (CDC22), held in Cancún, Mexico. This was joint work with Harsh Sharma and Boris Kramer at the University of California San Diego.
- November The group has received a DURIP award for "Graphics Processing Units based Computing System for Emerging Hypersonics Solvers." The PI for this project is Professor Venkat Raman and the other Co-I is Assistant Professor Aaron Towne. This project will fund a new computer system to develop advanced simulation and analysis algorithms that are tailored for GPU-based computing.
- November 17 Alex gave a talk at the Johns Hopkins University Civil and Systems Engineering Department Graduate Seminar. The title of the talk was "Multi-fidelity approaches to uncertainty quantification: sampling and surrogates"
- October 16–19 Alex presented at INFORMS 2022 in Indianapolis, Indiana, USA. at the Science-integrated Statistical Learning session. The title of his talk was "Compositional Multifidelity Models for Uncertainty Quantification via Directed Graphs. Slides can be found here.
- September 26–30 Nick and Doruk presented their recent research at the SIAM Conference on Mathematics of Data Science (MDS22) in San Diego, California. The title of his Nick's talk was "Enforcing Physical Structure in Bayesian Learning of Dynamical Systems: Stability and Energy Conservation," and Doruk's talk was "Incremental Tensor Train Decomposition for Compression and Approximation of High-Dimensional Streaming Datasets."
- August 18–19 Alex gave an invited talk on "Uncertainty Quantification with Multi-Fidelity Networks" at the USACM Thematic COnference on Uncertainty Quantification for Machine Learning Integrated Physics Modeling in Crystal City, Arlington, Virginia.
- June Josh presented his paper at the International Electric Propulsion Conference in Boston, MA. The title of his paper was "Simulation-based surrogate methodology of electric field for electrospray emitter geometry design and uncertainty quantification." In the paper we developed a neural network surrogate for predicting electric field strength for electrospray emitters and performed sensitivity analysis of emitter geometry.
- June Nick and Alex presented their recent research at the 8th European Congress on Computational Methods in Applied Sciences and Engineering (ECCOMAS Congress 2022) in Oslo, Norway. The title of Nick's talk was "Accounting for Model Uncertainty in the Identification of Partially Known Models." The title of Alex's talk was "Estimating Approximate Control Variate Weights: with Applications in Importance Sampling and Rare Event Simulation. Slides can be found here.
- May 2–6: Alex gives a presentation at the Workshop on Multilevel and multifidelity sampling methods in UQ for PDEs at the Erwin Schrödinger Institute (virtually): "MFNETS: Multifidelity surrogate modeling of ensembles of information sources through directed acyclic graphs." Slides can be found here.
- April: Nick, Doruk, Trung, and Alex presented their recent research at SIAM Conference on Uncertainty Quantification (UQ22) on topics ranging from Bayesian inference for dynamical systems, to deep learning for inverse design in dynamical systems to a new multi-fidelity surrogate approach to handle ensembles of information sources. Alex's slides can be found here.
- March Our paper on experimental design for multi-fidelity surrogate construction of multi-discplinary systems has been published in the International Journal of Numerical Methods in Engineering.
- January Hang's paper on a novel estimator for performing UQ in dynamical systems has been published in the Journal of Computational and Nonlinear Dynamics. The approach combines both surrogate and sampling approaches to extract that benefits of both of them while mitigating their downsides (slow convergence and low dimensionality)
- January Doruk's paper on a deep-learning approach to inverse design of self-oscillating sheets has been published in Neural Computing and Applications. This is joint work with Silas Alben and Robert Deegan at U-M.
- January Our paper with Micheline Soley, Paul Bergold, and Victor Batista on using the functional tensor train decomposition for quantum dynamics simulations has been published in the Journal of Chemical Theory and Computation.
- January Koray's paper on a novel approach to dynamic multi-agent assignment has now been published in the IEEE Transactions on Control of Networked Systems. This paper provides a new approach to assignment that is aware of agent dynamics. It is based on discrete optimal transport, applicable to systems with hundreds of linear or nonlinear agents, and highly scalable.
- January 3-7 We had two papers at AIAA Scitech this year. One on performing Bayesian inference for electrospray thrusters with the Plasmadynamics and Electric Propulsion Laboratory, it can be found here. The other paper was with NASA and Sandia National Laboratories on multifidelity uncertainty quantification for trajectory planning, it can be found here.
2021
- December Congratulations to Hang Yang for successfully defending his PhD! We wish him the best success in his upcoming endeavors.
- September the group welcomes two new graduate students Thomas Dixon and Joshua Eckels.
- August very well deserved congratulations to Brian and Nick for receiving the MICDE fellowship for research involving scientific computing and advancing computational discovery. Please see the press release here and here.
- August 17-20 Congratulations to Hang for achieving third place in the student paper competition at the 17th International Conference On Multibody Systems, Nonlinear Dynamics, and Control. His paper is titled "Multifidelity Uncertainty Quantification for Online Simulations of Automotive Propulsion Systems" and the paper can be found here. The paper demonstrates how intrusive polynomial chaos can be used in conjunction with approximate control variates to simultaneously reduce bias and variance in estimates of quantity of interest in dynamical systems.
- August 9-11 Alex presents work with Audelia Szulman, Collin Whittaker and Ben Jorns from Plasmadynamics and Electric Propulsion Laboratory at U-M at the AIAA Propulsion and Energy Forum. The paper "Robust Design of Electrospray Emitters" describes a way to generate robust feasible designs of electrospray thruster emitters and can be found here.
- June 28-July 1: Brian presents our paper "Behavioral Cloning in Atari Games Using a Combined Variational Autoencoder and Predictor Model" at the IEEE Congress on Evolutionary Computation. The paper can be found here here.
- June 28-30 Trung presents our work on approach for approximate control variates that is robust to mis-specified covariance matrices at the 4th International Conference on Uncertainty Quantification in Computational Sciences and Engineering. The paper can be found here.
- June 6-7: Nick presents a poster at the 3rd Annual Learning for Dynamics and Control Conference, hosted by ETH Zurich (remotely), describing a more robust objective for identifying linear systems from input-output data than traditional least-squares approaches. The paper can be found here.
- May 25-28: Alex presents at the Engineering Mechanics Institute Conference/Probabilistic Mechanics and Reliability 2021 Conference (EMI/PMC) on Bayesian system identification. Slides can be found here.
- May 7: Alex presents at the University of California, San Diego, Mechanical and Aerospace Engineering Dynamics and Control seminar. Slides can be found here.
- April 5: Hang receives the Distinguished Leadership Award from the College of Engineering. In the past couple of years, he has served multiple leadership roles in the College including co-chair of the 2021 Engineering Research Symposium (ERS), president of the Mechanical Engineering Graduate Council (MEGC), and a founding member of the Mechanical Engineering DEI committee.
- March 1-5: Trung and Nick present their recent research at SIAM Conference of Computational Science and Engineering (CSE21).
- March 3: We just released a preprint on inverse design of a self-oscillating gel using deep learning. This paper explores the construction of a deep learning architecture to map the motion of a sheet-like gel into parameters of its construction. You can find the paper, supplementary materials and videos, and the code at the following links preprint, supplementary material, code on bitbucket.
- February 25: We just released a preprint on reverse-mode differentiation for tensor networks with arbitrary network topologies. This paper shows how automatic differentiation schemes for sequences of tensor contractions can be enhanced by exploiting a few operation-specific properties. You can read the paper here.
- February 1: We just released a preprint on ensemble approximate control variate estimators. This paper addresses the issues regarding pilot sampling approaches for estimating correlations necessary to achieve variance reduction in multifidelity methods. You can read the paper here.
2020
- December 14-15 Alex presents at the Machine Learning in Science & Engineering Conference hosted by the Data Science Institute at Columbia about our Bayesian nonlinear system identification approaches. You can find the presentation here.
- December 14-18: Nick presents our paper "Bayesian Identification of Hamiltonian Dynamics from Symplectic Data" at CDC 2020 You can read the paper here.
- September 1: Nick's paper "Bayesian system ID: optimal management of parameter, model, and measurement uncertinty." was accepted to Nonlinear Dynamics. This paper provides a more robust formulation for learning dynamical systems from data than than a majority of least-squares based approaches currently used. You can read the paper here.
- September 1: Our collaborator John Jakeman made some really nice tutorials on our recently published MFNETS sampling framework for multifidelity UQ as part of the PyApprox package. You can find the tutorial here.
- July 1–3: Hang gives a presentation at the 2020 American Control Conference (virtually) on our paper "Uncertainty Quantification Using Generalized Polynomial Chaos for Online Simulations of Automotive Propulsion Systems."
- May 4–5: Alex gives a presentation at the Workshop on Multilevel and multifidelity sampling methods in UQ for PDEs at the Erwin Schrödinger Institute (virtually): "Sampling algorithms for generalized model ensembles in multifidelity uncertainty quantification." Slides can be found here.
- January 12–16: Alex gives a talk at the Physics Informed Machine Learning Workshop hosted by the Center for Nonlinear Studies at LANL, in Santa Fe, NM, USA: "Bayesian approaches for data-driven learning of dynamical systems." Slides can be found here.
2019
- October 15: Alex gives a talk at the Department of Mathematics at the University of Marlyand, College Park, MD, USA: "Compression algorithms for enabling high-dimensional motion planning." Slides can be found here.
- August 14-16: Alex gives a tutorial session at the 2019 UQ Summer School at USC in Los Angeles, CA, USA: "Tensor decompositions and their UQ applications". Slides can be found here.
- July 28 – August 1: Alex gives a talk at the 15th U.S. National Congress on Computational Mechanics in Austin, TX, USA: "Unifying multifidelity sampling approaches as approximate control variates"
- June 24–26: Alex gives a talk at the 3rd International Conference on Uncertainty Quantification in Computational Sciences and Engineering, in Crete, Greece: "New approaches to learn low-rank models of dynamical systems from streaming data."
- June 20: Alex gives an invited talk at LIMSI-CNRS, Orsay, France: "Compression algorithms for enabling high-dimensional motion planning"
- June 6 – 7: Alex gives a talk at the Physics Informed Machine Learning Workshop at the University of Washington, Seattle, WA, 2020: "Scalable learning of dynamical systems." Slides can be found here.
- May 23: Alex gives an invited talk at Sandia National Laboratories, Albuquerque, NM: "Compression algorithms for enabling high-dimensional motion planning"
2018
- 13th WCCM
- July 22–27: Alex gives a talk at the 13th World Congress in Computational Mechanics, New Your, NY, USA: "Multifidelity model management using latent variable networks"
- ACC2018
- June 27–29: Ezra Tal (MIT) gives a talk about our paper at ACC in Milwaukee, WI: "Continuous tensor train-based dynamic programming for high-dimensional zero-sum differential games"
- June 11–15: Alex gives a talk at the 6th European Conference on Computational Mechanics, in Glasgow, Scotland
- June 4–6: Alex gives a talk at the USC Workshop on Research Challenges and Opportunities at the interface of machine learning and uncertainty quantification, at USC, Los Angeles, CA, USA: "Regression in low-rank functional formats."
- April 16–19: Alex gives a talk at the SIAM Conference on Uncertainty Quantification in Orange County, CA, USA: "Sparse regularization for low-rank regression"
- January 30 – February 1: Alex participated in the DOE Workshop on Scientific Machine Learning in Rockville, MD.
- Jan 22: : Alex gives a talk at the Physics Constrained Machine Learning Workshop (hosted by the Center for Nonlienar Studies at LANL) in Santa Fe, NM, USA: "Multifidelity model management using latent variable Bayesian networks"
- January 19: Alex gives a talk in Ann Arbor at the Aerospace Graduate seminar: "Compression algorithms for enabling high-dimensional motion planning"
- January 18: Alex gives a talk in Ann Arbor at the Aerospace Graduate seminar: "Compression algorithms for enabling high-dimensional motion planning"
- January 1: Alex starts at University of Michigan
Get in Touch
- Call: 734-763-5214
- Check out our Github
- UM Aerospace Engineering