Formation and stability of spiking cell assemblies with spike-timing-dependent synaptic plasticity
FacultiesFakultät für Ingenieurwissenschaften und Informatik
It is commonly believed that our brains serve as information processing systems. Therefore, common methods used in computer science seem to be the first choice for investigating how these brains might work. But how is it possible to combine all the different parts of the brain which sometimes show chaotic behavior, and in the end it still seems to work in a coherent manner? How can this kind of stability emerge which guarantees us to always arrive at some decision point of a thought process? Most of the computational neuroscience community works on the level of neurons or small networks. In contrast, several theories try to exactly answer these questions. These so called brain theories mostly stay on a quite abstract level. One of these theories is named after its constituents: neural associative memories. But one aspect in this theory still does not get as much attention as recent experimental findings give reason to expect: learning. This is the starting point of this thesis. Not only should it be possible to learn new things, but information already stored in our brains should be resistant to fuzzy requests, i.e. fault tolerance makes this process even more complicated as it is already. Neurophysiology came up with some new findings in the last few years, especially regarding the topic of learning. In this thesis I applied this new class of learning algorithms/rules to networks of neural associative memories. One might argue that the choice of these learning rules seems somehow arbitrary. But it is not. Actually, these rules are accepted throughout all disciplines involved in brain research and even more important, leading scientists believe that they found the very basis of learning ubiquitous in our brains. So there are some good reasons why an accepted brain theory should be compatible with these rules. A mixture of empirical verifications by computer simulations and mathematical theory was used throughout this thesis.
Subject HeadingsAssoziativspeicher [GND]