IEEE CIS ACT Chapter Seminar: Computationally Expensive Optimization: A Surrogate Model-informed Design and Analysis Approach
IEEE ACT Virtual Seminar
Speaker: A/Prof. Pramudita Satria Palar, Faculty of Mechanical and Aerospace Engineering, Bandung Institute of Technology, Indonesia
Title: Computationally Expensive Optimization: A Surrogate Model-informed Design and Analysis Approach
==
Abstract: Computationally expensive optimization is ubiquitous in modern engineering design practice. Besides optimization, it is also valuable in other related tasks that include design exploration and uncertainty quantification, in which the latter leads to the notion of design under uncertainty. Surrogate models are indispensable tools in engineering design exploration that facilitate complex and expensive optimization. In this talk, some methods for efficient utilization of surrogate models to support engineering design optimization and analyses will be presented. The talk will primarily discuss the recent development and implementation of Gaussian Process Regression-based techniques for design optimization, exploration, and probabilistic analysis. Several approaches to incorporate auxiliary information such as derivatives and multi-fidelity simulation, to increase the efficiency of surrogate model, will also be discussed. The talk will also discuss improvements to Gaussian Process Regression through modifications of the covariance and trend functions, e.g., composite kernel and gradient-enhanced polynomial trend function. Finally, an example application on multi-objective optimization of a low-boom supersonic wing planform to simultaneously minimize drag and loudness will be presented.
==
Speaker biography: Pramudita Satria Palar is an Assistant Professor at the Faculty of Mechanical and Aerospace Engineering, Bandung Institute of Technology, Indonesia. Prior to his current position, he was a Research Fellow at the Institute of Fluid Science, Tohoku University. He completed his Ph.D. in Aeronautics and Astronautics from the University of Tokyo in 2015 and was a visiting researcher at the University of Cambridge in 2015 and at Leiden University in 2017. Dr. Palar's research interests are primarily in the areas of computationally expensive optimization; surrogate modeling, statistical / machine learning; uncertainty and sensitivity analysis; and their applications in aerospace, mechanical, and civil engineering. He has published several journal and refereed conference papers on the development of statistical learning-based techniques (primarily in Gaussian Process Regression) for design optimization and probabilistic design.
==
Date and Time
Location
Hosts
Registration
-
Add Event to Calendar
Loading virtual attendance info...
Speakers
A/Prof. Pramudita Palar
Computationally Expensive Optimization: A Surrogate Model-informed Design and Analysis Approach
Computationally expensive optimization is ubiquitous in modern engineering design practice. Besides optimization, it is also valuable in other related tasks that include design exploration and uncertainty quantification, in which the latter leads to the notion of design under uncertainty. Surrogate models are indispensable tools in engineering design exploration that facilitate complex and expensive optimization. In this talk, some methods for efficient utilization of surrogate models to support engineering design optimization and analyses will be presented. The talk will primarily discuss the recent development and implementation of Gaussian Process Regression-based techniques for design optimization, exploration, and probabilistic analysis. Several approaches to incorporate auxiliary information such as derivatives and multi-fidelity simulation, to increase the efficiency of surrogate model, will also be discussed. The talk will also discuss improvements to Gaussian Process Regression through modifications of the covariance and trend functions, e.g., composite kernel and gradient-enhanced polynomial trend function. Finally, an example application on multi-objective optimization of a low-boom supersonic wing planform to simultaneously minimize drag and loudness will be presented.
Biography:
Pramudita Satria Palar is an Assistant Professor at the Faculty of Mechanical and Aerospace Engineering, Bandung Institute of Technology, Indonesia. Prior to his current position, he was a Research Fellow at the Institute of Fluid Science, Tohoku University. He completed his Ph.D. in Aeronautics and Astronautics from the University of Tokyo in 2015 and was a visiting researcher at the University of Cambridge in 2015 and at Leiden University in 2017. Dr. Palar's research interests are primarily in the areas of computationally expensive optimization; surrogate modeling, statistical / machine learning; uncertainty and sensitivity analysis; and their applications in aerospace, mechanical, and civil engineering. He has published several journal and refereed conference papers on the development of statistical learning-based techniques (primarily in Gaussian Process Regression) for design optimization and probabilistic design.
Address:Indonesia