Sleep and Memory

Recent experimental and theoretical advances suggest that memories may be reorganized in the cortex during sleep. During sleep, our brains are highly active. The low-amplitude, high-frequency activity in the neocortex characteristic of the awake state is replaced with high amplitude, low-frequency rhythms during slow-wave sleep [1]. It would seem unlikely that the extensive corti-cal activity during sleep does not have some purpose; however, there is still no consensus on why we need to sleep. One intriguing possibility is that information acquired during the day is compared during sleep with older memories [2]. Previous neural network models included such a 'sleep phase' to calibrate the storage of memories acquired by Hebbian mechanisms [3-5]. Recent recordings from the hippocampus [6], and a new neural network model [7], lend experimental support and computational motivation to the possibility that we may sleep in order to organize efficient cortical representations of experience. Cortical representations of objects and events are widely distributed in the cerebral cortex. Thus, the representation of a violin might be stored in areas as diverse as the visual cortex, for its shape, the auditory cortex, for its sound, the parietal cortex, for how it may be grasped, and the motor cortex, for how it is played [8]. Problems arise when new experiences and objects must be integrated with existing information that is widely distributed. Learning algorithms designed for artificial neural networks that use such distributed representations can suffer from 'catastrophic interference' when new information is stored in the same neural circuits as old information [9]. Therefore, the brain must solve two problems during learning: where to make the changes needed to create a new memory; and how to make changes that are compatible with previously stored memories. There appears to be a period of consolidation before a memory becomes permanently stored. Thus, lesions of hippocampal formation, including the parahippocampal, perirhinal and entorhinal cortices, lead to memory deficits for up to 6 weeks following learning in monkeys [10], and more than a year in man. After this period of consolidation, lesions of the same areas are less disruptive, implying that the memories are stored elsewhere. Until recently, the processes that may occur in the cortex during the period of consolidation could only be inferred indirectly from such lesion experiments. Wilson and McNaughton [6] have reported changes that occur in the correlations between hippocampal neurons as a consequence of a new learning experience. They …