def custom_loss(y_true, y_pred):
    return K.sum(K.maximum(0, 1 - y_pred[0] + y_pred[1]))