Twin neural network regression is a semi-supervised regression algorithm

Abstract Twin neural network regression (TNNR) is trained to predict differences between the target values of two different data points rather than targets themselves. By ensembling predicted an unseen point and all training points, it possible obtain a very accurate prediction for original problem. Since any loop should sum zero, loops can be supplied data, even if themselves within are unlabelled. Semi-supervised improves TNNR performance, which already state art, significantly.