Fig. 16: Example explanation generating network.
Material attribute aware visual explanations (i.e., synthetic images) are generated from an image editing GAN129. The top row shows synthetic SEM images generated with increasing average particle size input. The lower row shows predictions from an experimental data trained material strength prediction model, in which the x-axis shows the normalized size attribute value. These prediction results agree with materials domain knowledge (i.e., the Hall-Petch rule). The material attribute aware image generation model together with the forward property prediction model form a general framework that can explain the effect of abstract material attributes. Figure reprinted from ref. 129 with permission. Copyright 2022 ACS.
