Deep-learning model using a small dataset

One the main challenges of developing data-driven models is the data-hungry nature of Artificial Neural Networks (ANNs). In our recent paper, we introduced a data augmentation approach to expand a small original dataset without conducting extra expensive high-fidelity simulations. We then used the original and augmented datasets for developing ANN models for non-linear path dependent composites. The obtained results showed the great impact of the data augmentation approach on the accuracy of the data-driven models. We believe this method has great potentials not only for other simulations and materials, but also for expanding experimental datasets. Here is the link (open access) to the paper: https://www.sciencedirect.com/science/article/pii/S0266353824000617