REM sleep appears to be a neural optimization and cleanup strategy. Here are a number of papers that've been written on the subject:
1987) The function of dreams (REM sleep): roles for the hippocampus, melatonin, monoamines, and vasotocin (Maurizi C.P.)
Rapid eye movement (REM) sleep is suggested to play a role in the storage of memory, resolution of emotional experiences, and erasure of memory (forgetting). Plasticity of hippocampal physiology, morphology, and chemistry seems to be evidence for new memory formation. REM sleep, melatonin, and monoamines may be involved in the transfer of memory from the intermediate-term high-capacity buffer in the hippocampus into long-term memory storage in the neocortex. Vasotocin, which is released by melatonin, could be an amnestic agent that erases recent memory from the hippocampal-entorhinal complex during dreams.
1988) REM sleep and neural nets (Crick F, Mitchison G.)
The broad features of rapid eye movement (REM) sleep are reviewed. Memory storage in the brain is probably quite unlike that in a digital computer, being distributed, superimposed and robust. Such memory systems are easily overloaded. If the stored memories share common features, random stimulation often produces mixed outputs. Simulations show that such overloading can be reduced by a process we call 'reverse learning'. We propose that this process is what is happening in REM sleep and that it explains in an unforced manner the condensation commonly found in dreams. Evidence for and against the proposed theory is discussed and several alternative theories are briefly described. The absence of REM sleep in the Enchidna and in two species of dolphins (that have relatively large brains) suggests that REM may allow the brain to be smaller than if REM were lacking.
1993) Crick and Mitchison's theory of REM sleep and neural networks (Brown D.W.)
Crick and Mitchison proposed that a reverse learning mechanism in REM sleep removes certain undesirable modes of interaction in neural networks within the cerebral cortex. If their theory is correct then abnormalities of reverse learning might account for some aspects of schizophrenia, mania, and depression. The theory of reverse learning might lead to an understanding of why lithium salts terminate manic episodes.
2001) Sleep, learning, and dreams: off-line memory reprocessing (Stickgold R, Hobson J.A, Fosse R, and Fosse M.)
Converging evidence and new research methodologies from across the neurosciences permit the neuroscientific study of the role of sleep in off-line memory reprocessing, as well as the nature and function of dreaming. Evidence supports a role for sleep in the consolidation of an array of learning and memory tasks. In addition, new methodologies allow the experimental manipulation of dream content at sleep onset, permitting an objective and scientific study of this dream formation and a renewed search for the possible functions of dreaming and the biological processes subserving it.