Title :
Topology preserving neural networks that achieve a prescribed feature map probability density distribution
Author :
Choi, Jongeun ; Horowitz, Roberto
Author_Institution :
Dept. of Mech. Eng., California Univ., Berkeley, CA, USA
Abstract :
In this paper, a new learning law for one-dimensional topology preserving neural networks is presented in which the output weights of the neural network converge to a set that produces a predefined winning neuron coordinate probability distribution, when the probability density function of the input signal is unknown and not necessarily uniform. The learning algorithm also produces an orientation preserving homeomorphic function from the known neural coordinate domain to the unknown input signal space, which maps a predefined neural coordinate probability density function into the unknown probability density function of the input signal. The convergence properties of the proposed learning algorithm are analyzed using the ODE approach and verified by a simulation study.
Keywords :
convergence; learning (artificial intelligence); probability; self-organising feature maps; topology; ODE approach; convergence properties; feature map probability density distribution; learning law; neural coordinate domain; one-dimensional neural networks; orientation preserving homeomorphic function; predefined winning neuron coordinate probability distribution; topology preserving neural networks; unknown input signal space; unknown probability density function; Convergence; Input variables; Mechanical engineering; Network topology; Neural networks; Neurons; Pattern recognition; Probability density function; Probability distribution; Signal processing algorithms;
Conference_Titel :
American Control Conference, 2005. Proceedings of the 2005
Print_ISBN :
0-7803-9098-9
Electronic_ISBN :
0743-1619
DOI :
10.1109/ACC.2005.1470151