Learning the action for long-time-step simulations of molecular dynamics