Path: utzoo!utgpu!news-server.csri.toronto.edu!rpi!zaphod.mps.ohio-state.edu!pacific.mps.ohio-state.edu!linac!att!pacbell.com!ucsd!sdcc6!beowulf!schraudo From: schraudo@beowulf.ucsd.edu (Nici Schraudolph) Newsgroups: comp.ai.philosophy Subject: Re: Dreams(Garbage collection) Message-ID: Date: 27 Mar 91 02:12:26 GMT References: <1991Mar25.114410.45892@vaxb.acs.unt.edu> Sender: news@sdcc6.ucsd.edu Lines: 35 munawar@vaxb.acs.unt.edu writes: >I was wondering whether dreams are our brain's method of >garbage collection(like done with computer memory) >Has anybody done any research in to it? Crick & Mitchison have speculated that the function of dream sleep might be to remove parasitic modes -- "false memories" that occur in some neural net architectures under heavy storage loads. The idea is that in REM sleep the brain "runs" in reverse, _unlearning_ from experience. Since there's no sensory input, experiences in REM sleep (ie. dreams) are dominated by the parasitic modes. This unlearning is a rough connectionist analog to garbage collection in symbolic computation, and is used in at least one neural net learning algorithm (Boltzmann Machine). Should it bear even the slightest resemblance to what happens in humans, dream content would offer fascinating insights into the cogs & gears of our cognitive machinery. Question: has this hypothesis by C&M had any impact on dream research? Have there been any developments in this area since they made it eight years ago? References: Crick & Mitchison (1983), The Function of Dream Sleep, Nature 304, 111-114. Hinton & Sejnowski (1986), Learning and Relearning in Boltzmann Machines, in: Rumelhart & McClelland (eds.), Parallel Distributed Processing, Vol. 1, 282-317, MIT Press. -- Nicol N. Schraudolph, CSE Dept. | "I don't know about your dreams, but mine Univ. of California, San Diego | are sort of hackney: same thing night after La Jolla, CA 92093-0114, U.S.A. | night, just this repetitive. And the color nici%cs@ucsd.{edu,bitnet,uucp} | is really bad..." - Laurie Anderson.