A5061119252- Model Reduction and Neural Networks
- Meteorological Phenomena and Simulations
- Gaussian Processes and Bayesian Inference
- Hydraulic and Pneumatic Systems
- Advanced Numerical Methods in Computational Mathematics
- Structural Health Monitoring Techniques
- Tribology and Lubrication Engineering
- Numerical methods for differential equations
- Climate variability and models
- Control Systems and Identification
- COVID-19 Pandemic Impacts
- Advanced Multi-Objective Optimization Algorithms
- Energy Load and Power Forecasting
- Data-Driven Disease Surveillance
- Probabilistic and Robust Engineering Design
- Fault Detection and Control Systems
- Plant Water Relations and Carbon Dynamics
- Computational Physics and Python Applications
- COVID-19 epidemiological studies
- Gear and Bearing Dynamics Analysis
- Advanced Combustion Engine Technologies
- Mechanical Failure Analysis and Simulation
- Mechanical Engineering and Vibrations Research
- Statistical and numerical algorithms
- Fluid Dynamics and Turbulent Flows
Columbia University
2023-2024
Environmental Earth Sciences
2023-2024
University of Pennsylvania
2020-2023
NSF National Center for Atmospheric Research
2023
Massachusetts Institute of Technology
2016-2021
We present a machine learning framework (GP-NODE) for Bayesian model discovery from partial, noisy and irregular observations of nonlinear dynamical systems. The proposed method takes advantage differentiable programming to propagate gradient information through ordinary differential equation solvers perform inference with respect unknown parameters using Hamiltonian Monte Carlo sampling Gaussian Process priors over the observed system states. This allows us exploit temporal correlations in...
This paper presents a machine learning framework for Bayesian systems identification from noisy, sparse and irregular observations of nonlinear dynamical systems. The proposed method takes advantage recent developments in differentiable programming to propagate gradient information through ordinary differential equation solvers perform inference with respect unknown model parameters using Hamiltonian Monte Carlo sampling. allows an efficient the posterior distributions over plausible models...
Modern climate projections lack adequate spatial and temporal resolution due to computational constraints. A consequence is inaccurate imprecise predictions of critical processes such as storms. Hybrid methods that combine physics with machine learning (ML) have introduced a new generation higher fidelity simulators can sidestep Moore's Law by outsourcing compute-hungry, short, high-resolution simulations ML emulators. However, this hybrid ML-physics simulation approach requires...
Body sensor networks are increasingly popular in healthcare, sports, military and security. However, the power supply from conventional batteries is a key bottleneck for development of body condition monitoring. Energy harvesting human motion to wearable or implantable devices promising alternative. This paper presents an airflow energy harvester harness footsteps. An air bladder-turbine designed convert footstep into electrical energy. The bladders embedded shoes induce foot-strikes....
Accurate and computationally-viable representations of clouds turbulence are a long-standing challenge for climate model development. Traditional parameterizations that crudely but efficiently approximate these processes leading source uncertainty in long-term projected warming precipitation patterns. Machine Learning (ML)-based have long been hailed as promising alternative with the potential to yield higher accuracy at fraction cost more explicit simulations. However, ML variants often...
Accurate representations of unknown and sub-grid physical processes through parameterizations (or closure) in numerical simulations with quantified uncertainty are critical for resolving the coarse-grained partial differential equations that govern many problems ranging from weather climate prediction to turbulence simulations. Recent advances have seen machine learning (ML) increasingly applied model these subgrid processes, resulting development hybrid physics-ML models integration...
This paper presents a machine learning framework for Bayesian systems identification from noisy, sparse and irregular observations of nonlinear dynamical systems. The proposed method takes advantage recent developments in differentiable programming to propagate gradient information through ordinary differential equation solvers perform inference with respect unknown model parameters using Hamiltonian Monte Carlo. allows us efficiently infer posterior distributions over plausible models...
A bstract This paper presents a deep learning framework for epidemiology system identification from noisy and sparse observations with quantified uncertainty. The proposed approach employs an ensemble of neural networks to infer the time-dependent reproduction number infectious disease by formulating tensor-based multi-step loss function that allows us efficiently calibrate model on multiple observed trajectories. method is applied mobility social behavior-based SEIR COVID-19 spread. trained...
Physical parameterizations (or closures) are used as representations of unresolved subgrid processes within weather and global climate models or coarse-scale turbulent models, whose resolutions too coarse to resolve small-scale processes. These typically grounded on physically based, yet empirical, the underlying Machine learning-based have recently been proposed an alternative solution shown great promise reduce uncertainties associated with parameterization Yet, those approaches still show...
Design and optimal control problems are among the fundamental, ubiquitous tasks we face in science engineering. In both cases, aim to represent optimize an unknown (black-box) function that associates a performance/outcome set of controllable variables through experiment. cases where experimental dynamics can be described by partial differential equations (PDEs), such mathematically translated into PDE-constrained optimization tasks, which quickly become intractable as number cost...
Interfacing challenges continue to impede the implementation of neural network-based parameterizations into numerical models atmosphere, particularly those written in Fortran. In this study, we leverage a specialized interfacing tool successfully implement parameterization for both deep and shallow convection within General Circulation Model, ARPEGE-Climat. Our primary objective is not only evaluate performance data-driven but also assess stability ARPEGE-Climat when coupled with trained on...
Abstract We present a simulation-based classification approach for large deployed structures with localized operational excitations. The method extends the two-level Port-Reduced Reduced-Basis Component (PR-RBC) technique to provide faster solution estimation hyperbolic partial differential equation of time-domain elastodynamics moving load. Time-domain correlation function-based features are built in order train classifiers such as Artificial Neural Networks and Support-Vector Machines...
Physical parameterizations are used as representations of unresolved subgrid processes within weather and global climate models or coarse-scale turbulent models, whose resolutions too coarse to resolve small-scale processes. These typically grounded on physically-based, yet empirical, the underlying Machine learning-based have recently been proposed an alternative shown great promises reduce uncertainties associated with Yet, those approaches still show some important mismatches that often...
Several fundamental problems in science and engineering consist of global optimization tasks involving unknown high-dimensional (black-box) functions that map a set controllable variables to the outcomes an expensive experiment. Bayesian Optimization (BO) techniques are known be effective tackling using relatively small number objective function evaluations, but their performance suffers when dealing with outputs. To overcome major challenge dimensionality, here we propose deep learning...
Machine-learning-based parameterizations (i.e. representation of sub-grid processes) global climate models or turbulent simulations have recently been proposed as a powerful alternative to physical, but empirical, representations, offering lower computational cost and higher accuracy. Yet, those approaches still suffer from lack generalization extrapolation beyond the training data, which is however critical projecting change unobserved regimes turbulence. Here we show that multi-fidelity...