And perhaps for good reason: We have long known that our artificial neural networks can benefit from periods of disconnected inputs where weight patterns can be tested for global conformance, but it seems now also that the brain is heavily involved in processing memories during 'slow-wave' sleep.

Of course, both theories threaten the psychiatric model of "dream interpretation" --- in these neuro-physiological models, dream activity seems to be throwing-out data, not acquiring new data. It could be that the bizarreness of our dreams is our brains trying to fit today's new experience data with what is already accepted, playing back alternate interpretations of the underlying sensory representations (memory is, remember, a holistic and active synthesis process, not a snapshot storage system) looking for patterns that will fit the existing 'accepted' interpretations.

It's worth noting that Iamblicus (c.200AD) was careful to distance the Egyptian priesthood from all dream interpretations ("The Mysteries of the Chaleans, Assyrians and Egyptians"), specifically identifying only the "lucid" dreams just after and just before consciousness as the only useful data for mystic interpretation.

Ooops ... Forgot the link!
http://dukemednews.org/news/article.php?id=7370