Filtering - Anealing : A two step hypothese for Sleep Information Management

D.CUGY, D.BASTARD, J.PATY

TheHebb rule, related to the synapse associative learning can be rephrased in a two parts rule as (1-2) :

• 1. If two neurons on either side of a synapse (connection) are activated simultaneously (i.e. synchronously), the strengh of that synapse is selectively increased.
• 2. If two neurons on either side of a synapse are activated asynchronously, that synapse is selectively weakened or eliminated.

During Slow Wave Sleep, the intra-hemispheric coherence of delta activity is increased relatively to the sleep stage (3). Phase ratio is described between centro-frontal and occipital EEG activity (4). High EEG coherence and phase ratio between EEG derivation let us propound than a forget function (characterized by the lowering of synapse weight) is associated to Slow Wave Sleep. This "forgeting function" acting as an Information Low Pass Filter i.e filtering all the weight information memorised during wake.

Simulated Anealing (5) is a process curently used for formal neural network learning optimisation. This process optimize the energy associated to the learned patterns. Denoyer & al (6) demonstrate a quick elevation (0,8°c) of cerebral temperature associated with REM Sleep followed by a slow descent. It is also known than neuronal activity is linked to temperature by Q10 factor. This direct link between temperature and neuronal activity let us propose than an anealing process is associated to Rem Sleep. Consequence of anealing is oversight and optimisation of knowledge.

(1) Stent G.S. A physiological mechanism for Hebb’s postulate for learning, 1973. Proc Nat Acad Science USA. 70,997-1001
(2) Changeux JP, Danchin A. Selective stabilization of developing synapses as a mechanism fo the specification of neural networks 1976. Nature 264, 705-712.
(3) Morvan C., Apports des Fonctions de Cohérence en Polysomnographie, Thèse Med Paris-Ouest 1991.
(4) Banquet J.P. Organisation spatio-temporelle de l’EEG des stades de sommeil. Rev EEG Neurophysiol, 11:75-81
(5) Harold Szu. Fast simulated annealing. In John S.Denker, editor, Neural Networks for Computing. American Institute of Physics, New-York, pages 420-425, 1986.
(5) Denoyer M., Sallanon M., Buda C., Delhomme G., Dittmar A., Jouvet M. The posterior hypothalamus is responsible for the increase of brain temperature during paradoxical sleep Exp. Brain Res. 84 (2) pp : 326-334 (1991)

Hebb rule, Neural Network, Slow Wave Sleep, REM Sleep, Anealing, Filtering, Oversight, Memory, Temperature.