Neural Network Algorithms

Minimal Resource Allocation Network: A sequential learning algorithm for realizing a minimal radial basis function network (RBFN), referred to as Minimal Resource Allocating Network (MRAN). With the growing and pruning strategy, the MRAN algorithm can implement a more compact network  structure, resulting in fast on-line learning. The MRAN algorithm is successfully applied in the areas of function approximation, time series prediction, nonlinear system identification and pattern recognition.

1. Yingwei Lu, N. Sundararajan, P. Saratchandran, “A Sequential Learning Scheme for Function Approximation using Minimal Radial Basis Function Neural Networks“, Neural Computation, MIT Press, USA, Vol. 9, No. 2, pp. 461-478, February 1997.

Matlab code: MRAN and EMRAN MATLAB Function Files

Communication Network

Call Admission Control: Minimal Resource Allocation Network (MRAN) and its extend version EMRAN is used to solve the traffic control problems in high-speed networks. Though the current focus is on the Call Admission Control (CAC) for Asynchronous Transfer Mode (ATM) networks, the method is applicable to any high-speed net-work. The models of MRAN and EMRAN CAC schemes have been designed by using OPNET. The network performances of MRAN and EMRAN CAC schemes are com-pared with the two conventional CAC schemes, Peak Band-width Allocation and Cell Loss Ratio (CLR) upper bound formula. Simulation results show that the new proposed CAC schemes are more efficient than the two conventional CAC approaches under the same traffic conditions.

Adaptive Flight Control

Until the later part of 1940's and early 1950s flight control research was aimed at providing pilot relief capabilities, mainly in the form of auto-pilots. With the advent of high performance aircrafts, it become evident that controllers were required to bring the aircrafts within certain specified operating envelop that would increase the pilots capabilities for controlling the aircrafts. Due to stringent requirements and complexities of the modern flight control system, it is difficult to estimate the nonlinearities accurately.  The performance of conventional controller will be poor under severe nonlinearities. The research deals with the new design development of adaptive and fault tolerant control systems using neural network methodologies. Conventional aircraft, helicopters and unmanned aerial vehicles are considered. The aim of the research area is the design of adaptive flight controllers such that the aircraft can even under severe fault and disturbances. The application of controllers to linear and nonlinear models was investigated.