@article{RabczukGuoZhuangetal.,
  author    = {Rabczuk, Timon and Guo, Hongwei and Zhuang, Xiaoying and Chen, Pengwan and Alajlan, Naif},
  title     = {Stochastic deep collocation method based on neural architecture search and transfer learning for heterogeneous porous media},
  series = {Engineering with Computers},
  volume    = {2022},
  journal   = {Engineering with Computers},
  publisher = {Springer},
  address   = {London},
  doi       = {10.1007/s00366-021-01586-2},
  url       = {http://nbn-resolving.de/urn:nbn:de:gbv:wim2-20220209-45835},
  pages     = {1 -- 26},
  abstract  = {We present a stochastic deep collocation method (DCM) based on neural architecture search (NAS) and transfer learning for heterogeneous porous media. We first carry out a sensitivity analysis to determine the key hyper-parameters of the network to reduce the search space and subsequently employ hyper-parameter optimization to finally obtain the parameter values. The presented NAS based DCM also saves the weights and biases of the most favorable architectures, which is then used in the fine-tuning process. We also employ transfer learning techniques to drastically reduce the computational cost. The presented DCM is then applied to the stochastic analysis of heterogeneous porous material. Therefore, a three dimensional stochastic flow model is built providing a benchmark to the simulation of groundwater flow in highly heterogeneous aquifers. The performance of the presented NAS based DCM is verified in different dimensions using the method of manufactured solutions. We show that it significantly outperforms finite difference methods in both accuracy and computational cost.},
  subject      = {Maschinelles Lernen},
  language  = {en}
}
@article{Aguinaga,
  author    = {Aguinaga, Jos{\´e} Guillermo De},
  title     = {Error in prediction due to data type availability in a coupled hydro-mechanical model},
  series = {Electronic Journal of Geotechnical Engineering},
  journal   = {Electronic Journal of Geotechnical Engineering},
  doi       = {10.25643/bauhaus-universitaet.3117},
  url       = {http://nbn-resolving.de/urn:nbn:de:gbv:wim2-20170413-31170},
  pages     = {2459 -- 2471},
  abstract  = {Different types of data provide different type of information. The present research analyzes the error on prediction obtained under different data type availability for calibration. The contribution of different measurement types to model calibration and prognosis are evaluated. A coupled 2D hydro-mechanical model of a water retaining dam is taken as an example. Here, the mean effective stress in the porous skeleton is reduced due to an increase in pore water pressure under drawdown conditions. Relevant model parameters are identified by scaled sensitivities. Then, Particle Swarm Optimization is applied to determine the optimal parameter values and finally, the error in prognosis is determined. We compare the predictions of the optimized models with results from a forward run of the reference model to obtain the actual prediction errors. The analyses presented here were performed calibrating the hydro-mechanical model to 31 data sets of 100 observations of varying data types. The prognosis results improve when using diversified information for calibration. However, when using several types of information, the number of observations has to be increased to be able to cover a representative part of the model domain. For an analysis with constant number of observations, a compromise between data type availability and domain coverage proves to be the best solution. Which type of calibration information contributes to the best prognoses could not be determined in advance. The error in model prognosis does not depend on the error in calibration, but on the parameter error, which unfortunately cannot be determined in inverse problems since we do not know its real value. The best prognoses were obtained independent of calibration fit. However, excellent calibration fits led to an increase in prognosis error variation. In the case of excellent fits; parameters' values came near the limits of reasonable physical values more often. To improve the prognoses reliability, the expected value of the parameters should be considered as prior information on the optimization algorithm.},
  subject      = {Sensitivit{\"a}tsanalyse},
  language  = {en}
}
@article{AmaniSaboorBagherzadehRabczuk,
  author    = {Amani, Jafar and Saboor Bagherzadeh, Amir and Rabczuk, Timon},
  title     = {Error estimate and adaptive refinement in mixed discrete least squares meshless method},
  series = {Mathematical Problems in Engineering},
  journal   = {Mathematical Problems in Engineering},
  doi       = {10.1155/2014/721240},
  url       = {http://nbn-resolving.de/urn:nbn:de:gbv:wim2-20170413-31181},
  abstract  = {The node moving and multistage node enrichment adaptive refinement procedures are extended in mixed discrete least squares meshless (MDLSM) method for efficient analysis of elasticity problems. In the formulation of MDLSM method, mixed formulation is accepted to avoid second-order differentiation of shape functions and to obtain displacements and stresses simultaneously. In the refinement procedures, a robust error estimator based on the value of the least square residuals functional of the governing differential equations and its boundaries at nodal points is used which is inherently available from the MDLSM formulation and can efficiently identify the zones with higher numerical errors. The results are compared with the refinement procedures in the irreducible formulation of discrete least squares meshless (DLSM) method and show the accuracy and efficiency of the proposed procedures. Also, the comparison of the error norms and convergence rate show the fidelity of the proposed adaptive refinement procedures in the MDLSM method.},
  subject      = {Elastizit{\"a}t},
  language  = {en}
}