Acetylcholine and learning in a cortical associative memory

• Authors: Michael E. Hasselmo
• Venue: Neural Computation
• Year: 1993

Implementing associative memory function in biologically realisticnetworks raises difficulties not dealt with in previous associativememory models. In particular, during learning of overlapping inputpatterns, recall of previously stored patterns can interfere withthe learning of new patterns. Most associative memory models avoidthis difficulty by ignoring the effect of previously modifiedconnections during learning, thereby clamping activity to thepatterns to be learned. Here I propose that the effects ofacetylcholine in cortical structures may provide aneuropsychological mechanism for this clamping. Recent brain sliceexperiments have shown that acetylcholine selectively suppressesexcitatory intrinsic fiber synaptic transmission within theolfactory cortex, while leaving excitatory afferent inputunaffected. In a computational model of olfactory cortex, thisselective suppression, applied during learning, preventsinterference from previously stored patterns during the learning ofnew patterns. Analysis of the model shows that the amount ofsuppression necessary to prevent interference depends on corticalparameters such as inhibition and the threshold of synapticmodification, as well as input parameters such as the amount ofoverlap between the patterns being stored.