The main goal of this thesis has been to evaluate, both quantitativily and qualitativily, to what degree it is possible to get a computer model of Bayesian incremental learning behave like experimental physiological data regarding LTP, long-term potentiation. Especially when there is temporal differences in the stimulus to the neurons. By adjusting the initial probability values it is possible to simulate various previous stimuli of the neurons. Changing the time constants allows the synapses weight changes to partly resemble biological data, a small or neglectible change when the time difference is large, and a potentiation when the presynaptic stimuli arrives before the postsynaptic one. It was not possible to recreate the clearly appearant depression when the postsynaptic stimuli arrives before presynaptic one.
View preliminary report in postscript(1.1MB)