Yet another approach is taken by the BERKMIN algorithm
[Goldberg NovikovGoldberg Novikov2002]. Here, the set of nogoods is partitioned into two
separate groups based on how recently the nogoods were acquired;
of the nogoods are kept in a ``recent'' group and the
remaining
in an ``old'' group. A relatively large
length bound is used to cull the recently acquired nogoods while a
smaller length bound is used to aggressively cull the smaller group of
older nogoods. We are not aware of any studies comparing these hybrid
approaches to pure length-bounded or relevance-bounded methods.