The abstract seems to be comparing having to search an entire problem space for the most optimal result ("the current approach") with an approach using sampling for a good enough result?