Known Unknowns: Out-of-Distribution Property Prediction in Materials

Developing high-performance materials often requires identifying materials with property values that lie outside the known distribution. Therefore, the ability to extrapolate to out-of-support material property values is invaluable to materials design. Given chemical compositions and their property values, our objective is to learn a predictor that extrapolates zero-shot to higher ranges. In this work, we employ a transductive approach to property prediction and explore its extrapolation capabilities. We leverage analogical composition-target relations in the training and test sets, enabling generalization beyond the training target support.