IIIT Hyderabad Publications |
|||||||||
|
Linear Congruential Sequences : Feedback and Recurrent Neural NetworksAuthors: Garimella Ramamurthy Date: 2015-02-04 Report no: IIIT/TR/2015/7 AbstractIn this research paper, state space representation of a non-linear dynamical system associated with a linear congruential sequence is discussed. Based on the periodicity of such sequence, it is inferred that the dynamical system exhibits cycles in the state space. The cycle length is determined. A novel model of neuron, called “modulo” neuron is proposed. Based on such a neuron, an associative memory and certain recurrent neural networks are proposed. Also, a lower bound on the largest eigenvalue of non-negative, symmetric synaptic weight matrix is derived ( thus lower bounding the maximum energy value ). Full report: pdf Centre for Communications |
||||||||
Copyright © 2009 - IIIT Hyderabad. All Rights Reserved. |