Computational capacity of time-recurrent networks
Sergei Avgustovich Vakulenko · Journal of Physics A Mathematical and General · 2002
Time-recurrent networks are considered. Synaptic plasticity is defined by a simple Hebb rule. It is well known that this Hebbian mechanism can support learning and memory. We show that this plasticity is a computational instrument with large possibilities. In particular, the synaptic matrix can store different information, both dynamic and static. For example, the network can perform the Fourier and wavelet transformations and calculate probability distributions of unknown parameters. These networks can analyse and identify dynamics, calculate likelihood, study autoregression etc. They can resolve even more sophisticated problems, for example decoding fractal images.