In case of overfitting (using val loss) the last and the min might not be even near, but maybe hyperparam aborts in those cases? I am not too familiar on when hyperparam optmizer does abort an experiment