Improving neural network efficiency with multifidelity and dimensionality reduction techniques
(2025) AIAA SCITECH 2025 Forum- Abstract
- Design problems in aerospace engineering often require numerous evaluations of expensive-to-evaluate high-fidelity models, resulting in prohibitive computational costs. One way to address the computational cost is through building surrogates, such as deep neural networks (DNNs). However, DNNs may only be an effective surrogate when sufficient evaluations of the high-fidelity model are required such that the up-front training cost is amortized, or in situations that require real-time responses (such as interactive visualizations). Typically, the data requirements for adequately accurate training of DNNs are often impractical for engineering applications. To alleviate this issue, the proposed work utilizes output dimensionality reduction... (More)
- Design problems in aerospace engineering often require numerous evaluations of expensive-to-evaluate high-fidelity models, resulting in prohibitive computational costs. One way to address the computational cost is through building surrogates, such as deep neural networks (DNNs). However, DNNs may only be an effective surrogate when sufficient evaluations of the high-fidelity model are required such that the up-front training cost is amortized, or in situations that require real-time responses (such as interactive visualizations). Typically, the data requirements for adequately accurate training of DNNs are often impractical for engineering applications. To alleviate this issue, the proposed work utilizes output dimensionality reduction along with information from multiple models of varying fidelities and cost to develop accurate projection-enabled multifidelity neural networks (MF-NNs) with limited training samples. The dimensionality reduction leads to a more parsimonious network and the multifidelity aspect adds more training data from lower-cost, lower-fidelity models. Three approaches for MF-NNs that leverage proper orthogonal decomposition based projections are introduced: (i) pre-training method, (ii) additive method, and (iii) multi-step method. The MF-NN is applied to approximate the optimal design of 2D aerodynamic airfoils given the performance and design requirements. The MF-NN leads to ~27% computational cost reduction compared to single-fidelity neural networks at the same accuracy (90%), with the multi-step approach performing the best for this application. (Less)
Please use this url to cite or link to this publication:
https://lup.lub.lu.se/record/2a3dcb9d-82de-4afb-9eba-6d215286ac3e
- author
- Sella, Vignesh ; O’Leary-Roseberry, Thomas ; Du, Xiaosong ; Guo, Mengwu LU ; Martins, Joaquim R. R. A. ; Ghattas, Omar ; Willcox, Karen and Chaudhuri, Anirban
- organization
- publishing date
- 2025-01-03
- type
- Chapter in Book/Report/Conference proceeding
- publication status
- published
- subject
- host publication
- AIAA SciTech 2025 Forum
- publisher
- American Institute of Aeronautics and Astronautics
- conference name
- AIAA SCITECH 2025 Forum
- conference location
- Orlando, Florida, United States
- conference dates
- 2025-01-06 - 2025-01-10
- external identifiers
-
- scopus:105001330458
- ISBN
- 9781624107238
- DOI
- 10.2514/6.2025-2807
- language
- English
- LU publication?
- yes
- id
- 2a3dcb9d-82de-4afb-9eba-6d215286ac3e
- date added to LUP
- 2025-01-04 19:34:45
- date last changed
- 2025-06-10 04:01:18
@inproceedings{2a3dcb9d-82de-4afb-9eba-6d215286ac3e,
abstract = {{Design problems in aerospace engineering often require numerous evaluations of expensive-to-evaluate high-fidelity models, resulting in prohibitive computational costs. One way to address the computational cost is through building surrogates, such as deep neural networks (DNNs). However, DNNs may only be an effective surrogate when sufficient evaluations of the high-fidelity model are required such that the up-front training cost is amortized, or in situations that require real-time responses (such as interactive visualizations). Typically, the data requirements for adequately accurate training of DNNs are often impractical for engineering applications. To alleviate this issue, the proposed work utilizes output dimensionality reduction along with information from multiple models of varying fidelities and cost to develop accurate projection-enabled multifidelity neural networks (MF-NNs) with limited training samples. The dimensionality reduction leads to a more parsimonious network and the multifidelity aspect adds more training data from lower-cost, lower-fidelity models. Three approaches for MF-NNs that leverage proper orthogonal decomposition based projections are introduced: (i) pre-training method, (ii) additive method, and (iii) multi-step method. The MF-NN is applied to approximate the optimal design of 2D aerodynamic airfoils given the performance and design requirements. The MF-NN leads to ~27% computational cost reduction compared to single-fidelity neural networks at the same accuracy (90%), with the multi-step approach performing the best for this application.}},
author = {{Sella, Vignesh and O’Leary-Roseberry, Thomas and Du, Xiaosong and Guo, Mengwu and Martins, Joaquim R. R. A. and Ghattas, Omar and Willcox, Karen and Chaudhuri, Anirban}},
booktitle = {{AIAA SciTech 2025 Forum}},
isbn = {{9781624107238}},
language = {{eng}},
month = {{01}},
publisher = {{American Institute of Aeronautics and Astronautics}},
title = {{Improving neural network efficiency with multifidelity and dimensionality reduction techniques}},
url = {{http://dx.doi.org/10.2514/6.2025-2807}},
doi = {{10.2514/6.2025-2807}},
year = {{2025}},
}