On Adam Trained Models and a Parallel Method to Improve the Generalization PerformanceGuojing CongLuca Buratti2018MLHPC 2018