The Hebbian hypothesis of activity-dependent synaptic plasticity has gained much support from experimental studies of long-term potentiation and depression. Such studies have also uncovered complex patterns of competition among the synapses. Such effects may be due to the neuron redistributing its limited synaptic resources as synaptic strengths change. In computational models this strategy is commonly known as normalized Hebbian learning. However, not much consideration is usually given to whether the weights are normalized over the presynaptic or the postsynaptic sites of the neuron. Our results show that the different loci of normalization can result in drastic differences in the model's behavior, suggesting that future experiments should investigate presynaptic factors of redistribution as well as the more widely studied postsynaptic factors.

PDF, PS

Neurocomputing, 32--33:77-82, 2000.

@Article{choe:cns99, title={Effects Of Presynaptic And Postsynaptic Resource Redistribution In Hebbian Weight Adaptation}, author={Yoonsuck Choe and Risto Miikkulainen and Lawrence K. Cormack}, volume={32--33}, journal={Neurocomputing}, pages={77-82}, url="http://nn.cs.utexas.edu/?choe:neurocomputing00", year={2000} }