Deep Adaptive Arbitrary Polynomial Chaos Expansion: A Mini-data-driven Semi-supervised Method for Uncertainty Quantification

Wen Yao,Xiaohu Zheng,Jun Zhang,Ning Wang,Guijian Tang
DOI: https://doi.org/10.48550/arXiv.2107.10428
2022-03-01
Abstract:The surrogate model-based uncertainty quantification method has drawn much attention in many engineering fields. Polynomial chaos expansion (PCE) and deep learning (DL) are powerful methods for building a surrogate model. However, PCE needs to increase the expansion order to improve the accuracy of the surrogate model, which causes more labeled data to solve the expansion coefficients, and DL also requires a lot of labeled data to train the deep neural network (DNN). First of all, this paper proposes the adaptive arbitrary polynomial chaos (aPC) and proves two properties about the adaptive expansion coefficients. Based on the adaptive aPC, a semi-supervised deep adaptive arbitrary polynomial chaos expansion (Deep aPCE) method is proposed to reduce the training data cost and improve the surrogate model accuracy. For one hand, the Deep aPCE method uses two properties of the adaptive aPC to assist in training the DNN based on only a small amount of labeled data and many unlabeled data, significantly reducing the training data cost. On the other hand, the Deep aPCE method adopts the DNN to fine-tune the adaptive expansion coefficients dynamically, improving the Deep aPCE model accuracy with lower expansion order. Besides, the Deep aPCE method can directly construct accurate surrogate models of the high dimensional stochastic systems without complex dimension-reduction and model decomposition operations. Five numerical examples and an actual engineering problem are used to verify the effectiveness of the Deep aPCE method.
Machine Learning
What problem does this paper attempt to address?
The main problem that this paper attempts to solve is to improve the accuracy of surrogate models in uncertainty quantification while reducing the number of samples required to construct these models. Specifically, the paper proposes the Adaptive Polynomial Chaos Expansion (aPC) method and, on this basis, develops a semi - supervised Deep Adaptive Polynomial Chaos Expansion (Deep aPCE) method. This method aims to overcome the "curse of dimensionality" problem and the need for a large amount of labeled data when traditional Polynomial Chaos Expansion (PCE) and deep - learning methods deal with high - dimensional stochastic systems by combining the advantages of deep learning (DL) and adaptive aPC. The paper points out that in order to improve the model accuracy, the traditional PCE method needs to increase the expansion order, which will lead to an increase in the number of samples required to solve the expansion coefficients, thereby increasing the computational cost. Although deep learning can construct high - accuracy surrogate models, it also requires a large amount of labeled training data to train the Deep Neural Network (DNN), which will also lead to unacceptable computational costs. Therefore, the method proposed in the paper aims to use a small amount of labeled data and a large amount of unlabeled data to train the DNN, significantly reducing the cost of training data while improving the accuracy of the surrogate model. In addition, the Deep aPCE method can directly construct accurate surrogate models for high - dimensional stochastic systems without complex dimensionality reduction or model decomposition operations. By dynamically adjusting the adaptive expansion coefficients, this method can further improve the model accuracy. Overall, the method proposed in the paper aims to provide an efficient and accurate solution for uncertainty quantification, especially suitable for the analysis of complex stochastic systems in the engineering field.