Independent component analysis, a new concept?
Signal Processing - Special issue on higher order statistics
Stability and Chaos of a Class of Learning Algorithms for ICA Neural Networks
Neural Processing Letters
Dynamic behavior of a discrete modified Ricker & Beverton-Holt model
Computers & Mathematics with Applications
IEEE Transactions on Signal Processing
On the discrete-time dynamics of the basic Hebbian neural network node
IEEE Transactions on Neural Networks
Convergence analysis of a deterministic discrete time system of Oja's PCA learning algorithm
IEEE Transactions on Neural Networks
Hi-index | 0.09 |
The stability and chaos of an ICA algorithm are investigated by analyzing the corresponding deterministic discrete time (DDT) system. The existence and stability of all the possible fixed points of the ICA algorithm are studied. While the nonlinear function contained in the algorithm is specified, an invariant set of the algorithm is obtained so that the non-divergence of the algorithm can be guaranteed. It is then derived in this invariant set that the behaviors of the algorithm are dominated by a one-dimensional map. The conditions for convergence and chaos are derived. In the outside of the invariant set, the corrected Marotto's theorem and computer-assisted method are applied to study the two-dimensional case of the algorithm and the existence of chaos is proved. The attractors and bifurcation diagrams of the algorithm with different parameters are presented to further confirm the obtained results.