Four datasets for multi-input convolutional network

Datasets were produced from massive (multi-)physics simulations. They are used to train multi-input convolutional network, which then can act as a cheap substitute of original physics-based models and allows for ultrafast simulation.

The datasets and four related physical and engineering problems have distinct characteristics, which should present different challenges to a multi-input ConvNet. They can help comprehensively test the modeling capability of a multi-input ConvNet. Note that the data requires further processing, namely properly preparing multi-input-output pairs, i.e.,((a,X), Y), for training the multi-input convolutional network.