A5080481874- Model Reduction and Neural Networks
- Numerical methods for differential equations
- Control Systems and Identification
- Probabilistic and Robust Engineering Design
- Real-time simulation and control systems
- Machine Learning in Materials Science
- Hydraulic and Pneumatic Systems
- Topic Modeling
- Power System Optimization and Stability
- Complex Network Analysis Techniques
- X-ray Diffraction in Crystallography
- Fluid Dynamics and Vibration Analysis
- Neural Networks and Applications
- Stock Market Forecasting Methods
- Advanced Materials Characterization Techniques
- Structural Health Monitoring Techniques
- Control and Stability of Dynamical Systems
- Fluid Dynamics and Turbulent Flows
- Nuclear Engineering Thermal-Hydraulics
- Fault Detection and Control Systems
- Advanced Graph Neural Networks
- Gaussian Processes and Bayesian Inference
- Scientific Computing and Data Management
- Advanced Control Systems Optimization
- Electron and X-Ray Spectroscopy Techniques
Max Planck Institute for Dynamics of Complex Technical Systems
2015-2024
Indian Institute of Technology Kharagpur
2020-2023
Max Planck Society
2014-2021
Weierstrass Institute for Applied Analysis and Stochastics
2021
Martin Luther University Halle-Wittenberg
2021
Jaipur Golden Hospital
2020
University of Ulster
2009
The University of Texas at Austin
1994-1997
Abstract The purpose of this work is the development a trained artificial neural network for surrogate modeling mechanical response elasto-viscoplastic grain microstructures. To end, U-Net-based convolutional (CNN) using results von Mises stress field from numerical solution initial-boundary-value problems (IBVPs) equilibrium in such microstructures subject to quasi-static uniaxial extension. resulting CNN (tCNN) accurately reproduces about 500 times faster than solutions corresponding IBVP...
Discovering dynamical models to describe underlying behavior is essential draw decisive conclusions and engineering studies, e.g., optimizing a process. Experimental data availability notwithstanding has increased significantly, but interpretable explainable in science yet remain incomprehensible. In this work, we blend machine learning dictionary-based with numerical analysis tools discover governing differential equations from noisy sparsely-sampled measurement data. We utilize the fact...
In this paper, we define a class of generalized guaranteed rate (GR) scheduling algorithms that includes which allocate variable to the packets flow. We work-conserving virtual clock, packet-by-packet processor sharing, and self-clocked fair queueing can also suitable for servers where packet fragmentation may occur. demonstrate if controllers is employed flow in conjunction with any algorithm GR, then resulting non-work-conserving belongs GR. This leads definition several algorithms....
We investigate the optimal model reduction problem for large-scale quadratic-bilinear (QB) control systems. Our contributions are threefold. First, we discuss variational analysis and Volterra series formulation QB then define $\mathcal H_2$-norm a system based on kernels of underlying propose truncated as well. Next, derive first-order necessary conditions an approximation, where optimality is measured in terms error system. iterative algorithm, which upon convergence yields reduced-order...
Abstract Science is and always has been based on data, but the terms ‘data-centric’ ‘4th paradigm’ of materials research indicate a radical change in how information retrieved, handled performed. It signifies transformative shift towards managing vast data collections, digital repositories, innovative analytics methods. The integration artificial intelligence its subset machine learning, become pivotal addressing all these challenges. This Roadmap Data-Centric Materials explores fundamental...
Scientific machine learning for inferring dynamical systems combines data-driven modeling, physics-based and empirical knowledge. It plays an essential role in engineering design digital twinning. In this work, we primarily focus on operator inference methodology that builds models, preferably low-dimension, with a prior hypothesis the model structure, often determined by known physics or given experts. Then, inference, aim to learn operators of setting up appropriate optimization problem....
NLP in the legal domain has seen increasing success with emergence of Transformer-based Pre-trained Language Models (PLMs) pre-trained on text. PLMs trained over European and US text are available publicly; however, from other domains (countries), such as India, have a lot distinguishing characteristics. With rapidly volume Legal applications various countries, it become necessary to pre-train LMs countries well. In this work, we attempt investigate pre-training Indian domain. We re-train...
Proposes a novel observation-based admission control algorithm, in which client is admitted for service by multimedia server only if the predicted extrapolation from status quo measurements of storage utilization indicate that requirements all clients can be met satisfactorily. The performance and hence, number serviced simultaneously are maximized employing disk scheduling algorithm minimizes both seek rotational latency incurred while accessing sequence media blocks disk. effectiveness...
In this work, we investigate a model-order reduction scheme for polynomial systems. We begin with defining the generalized multivariate transfer functions system. Based on this, aim at constructing reduced-order system, interpolating defined given set of interpolation points. Furthermore, provide method, inspired by Loewner approach linear and (quadratic-)bilinear systems, to determine good-quality system in an automatic way. also discuss computational issues related proposed method...
Measurement noise is an integral part of collecting data a physical process. Thus, removal necessary to draw conclusions from these data, and it often becomes essential construct dynamical models using data. We discuss methodology learn differential equation(s) noisy irregularly sampled measurements. In our methodology, the main innovation can be seen in integration deep neural networks with ordinary equations (ODEs) approach. Precisely, we aim at learning network that provides...
We present a framework for learning Hamiltonian systems using data. This work is based on lifting hypothesis, which posits that nonlinear can be written as with cubic Hamiltonians. By leveraging this, we obtain quadratic dynamics are in transformed coordinate system. To end, given generalized position and momentum data, propose methodology to learn dynamical systems, enforcing the structure combination weakly-enforced symplectic autoencoder. The obtained exhibits long-term stability of...
We discuss balanced truncation model order reduction for large-scale quadratic-bilinear (QB) systems. Balanced linear systems mainly involves the computation of Gramians system, namely reachability and observability Gramians. These are extended to a general nonlinear setting in Scherpen (1993), where it is shown that solutions state-dependent Hamilton-Jacobi equations. Therefore, they not only difficult compute but also hard utilize framework. In this paper, we propose algebraic QB based on...
Abstract We present a deep-learning framework, CrysXPP, to allow rapid and accurate prediction of electronic, magnetic, elastic properties wide range materials. CrysXPP lowers the need for large property tagged datasets by intelligently designing an autoencoder, CrysAE. The important structural chemical captured CrysAE from amount available crystal graphs data helped in achieving low errors. Moreover, we design feature selector that helps interpret model’s prediction. Most notably, when...
The engineering design process often relies on mathematical modeling that can describe the underlying dynamic behavior. In this work, we present a data-driven methodology for dynamics of nonlinear systems. To simplify task, aim to identify coordinate transformation allows us represent systems using common, simple model structure. advantage common is customized tools developed it be applied study large variety simplest model—one think — linear, but linear fall short in accurately capturing...
Abstract Discovering a suitable coordinate transformation for nonlinear systems enables the construction of simpler models, facilitating prediction, control, and optimization complex systems. To that end, Koopman operator theory offers framework global linearization systems, thereby allowing usage linear tools design studies. In this work, we focus on identification linearized embeddings canonical Hamiltonian through symplectic transformation. While task is often challenging, leverage power...
Reduced-order modeling has a long tradition in computational fluid dynamics. The ever-increasing significance of data for the synthesis low-order models is well reflected recent successes data-driven approaches such as Dynamic Mode Decomposition and Operator Inference. With this work, we discuss an approach to learning structured incompressible flow from that can be used engineering studies control, optimization, simulation. To end, utilize intrinsic structure Navier-Stokes equations flows...
Abstract Linear projection schemes like Proper Orthogonal Decomposition can efficiently reduce the dimensions of dynamical systems but are naturally limited, e.g., for convection‐dominated problems. Nonlinear approaches have shown to outperform linear methods in terms dimension reduction versus accuracy but, typically, come with a large computational overhead. In this work, we consider quadratic scheme which induces nonlinear structures that well accessible tensorized algebra routines. We...