作者: Richard Bowles
DOI:
关键词:
摘要: Cell assemblies are co-operating groups of neurons believed to exist in the brain. Their existence was proposed by neuropsychologist D.O. Hebb who also formulated a mechanism which they could form, now known as Hebbian learning. Evidence for learning and cell brain is accumulating investigation tools improve. Researchers have simulated neural networks computers. This thesis describes simulations assemblies. The feasibility that possess all predicted properties biological established. can be coupled together with weighted connections form hierarchies group basic assemblies, termed primitives connected such way compound assembly. component these ignited independently, i.e. activated due signals being passed entirely within network, but if sufficient number them. activated, co-operate ignite remaining Various experiments described subject external activation involving cells those stimulated artificially high level. These then fire, produce spike activity analogous spiking neurons, this pass their other cells. Connections established, some set others, between different ones, allow from one primitive another. In way, activating or more may cause others ignite. Experiments spontaneous aids recruitment uncommitted neighbouring strong relationship Hopfield nets described. A network support numbers depending on complexity Assemblies classified terms how many present each assembly minimum needed complete it. A 2-3 contains 3 primitives, any 2 will N hold order an architecture O(N2) 3-4 show emanating must scaled up linearly .increases maintain same mean primitive. Restricting maximum leads, severe loss performance size increases. It shown duplicated nets, there restrictions carrying capacity either hierarchy net storing patterns, promise N2 patterns largely illusory. When fixed increased, only O(N) stored.