Reutilization of Rejected Training Data for Taming Learning Problem