Data-driven methods are of increasing popularity for solving problems in geotechnics, offering as they do, the possibility of high-fidelity results without the effort of a detailed deterministic numerical analysis (e.g. using finite elements). A wide range of approaches fall under the heading of Reduced Order Models (ROMs) which are created by processing data generated from high-fidelity models. The quality of these ROMs, and the computational cost of their construction, themselves depend heavily on the architecture chosen. In this study, we introduce a set of efficient frameworks for data-driven ROMs that can be applied to geotechnics problems in general. Our approach employs autoencoders and/or principal component analysis to reduce data dimensionality and to extract latent representations, followed by a Deep Operator Network (DeepONet) to learn nonlinear behaviour within this latent space. The architectures are demonstrated on the problem of the prediction of spatio-temporal responses in soil consolidation, and we demonstrate that the proposed efficient ROM architectures accurately predict responses for a range of problem specifications. The proposed framework provides a versatile methodology for large-scale complex geotechnical modelling applications.
Mao Ouyang, C. Augarde, W. Coombs et al.· Acta Geotechnica· 0 citations
The process of selecting suitable clients to participate in Federated Learning (FL) remains a critical challenge due to non-IID heterogeneity in data distributions and heterogeneity in computational resources among participating clients. Existing client selection approaches typically assess clients individually using attributes such as model accuracy. Such individual-based client assessment ignores group effects, which are paramount for global model performance. In this paper, we propose a Genetic-Algorithm (GA) based client selection mechanism that is applicable to both horizontal FL (HFL) and vertical FL (VFL). Candidate client groups are evaluated by a multi-criteria fitness function that jointly models group data size, feature coverage, label coverage, class balance, feature-distribution divergence, feature importance, computational power, reputation, accuracy, and outlier proportion. The GA performs an efficient search over the combinatorial space of client subsets. The proposed approach is evaluated using image classification, diabetes prediction, and rain prediction tasks. The proposed approach improves global model accuracy and accelerates convergence. On the MNIST dataset, it peaks at 98.37%, outperforming DSCS (97.27%) and FedMCCS (97.07%). On the diabetes dataset, it achieves 88.37%, while DSCS and FedMCCS achieve 87.15% and 86.64%, respectively. On the rain prediction task, it attains a peak accuracy of 83.57% and converges to 83.36%, compared with 81.98% for FedMCCS and 80.38% for DSCS.
Sani Umar, Ahmed Alagha, R. Mizouni et al.· Evolutionary Intelligence· 0 citations
