Abstract
Differentially activated areas of a dendrite permit the existence of zones with distinct rates of synaptic modification, and such areas can be individually accessed using a reference signal which localizes synaptic plasticity and memory trace retrieval to certain subregions of the dendrite. It is proposed that the neural machinery required in such a learning/retrieval mechanism could involve the NMDA receptor, in conjunction with the ability of dendrites to maintain differentially activated regions. In particular, it is suggested that such a parcellation of the dendrite allows the neuron to participate in multiple sequences, which can be learned without suffering from the 'wash-out' of synaptic efficacy associated with superimposition of training patterns. This is a biologically plausible solution to the stability-plasticity dilemma of learning in neural networks.
Original language | English |
---|---|
Journal | Network-computation in Neural Systems |
Volume | 14 |
Issue number | 2 |
Pages (from-to) | 189-209 |
ISSN | 0954-898X |
Publication status | Published - 2003 |