| || Missile Defence and Interceptor Allocation by LVQ-RBF Multi-agent Hybrid Architecture
Author : Thamarai Selvi, S.;Malmathanraj, R.
Source : Defence Science Journal ; Vol:57(2) ; 2007 ; pp 173-183
Subject : 629.7 Aeronautics
Keywords : Reinforcement learning;LVQ neural network;Theatre missile defence;Interceptor allocation;Q-learning;RBF neural network;Radial basis function;Learning vector quantisation
Abstract : This paper proposes a solution methodology for a missile defence problem using theatre missile defence (TMD) concept. In the missile defence scenario, the concept of TMD is generally used for the optimal allocation of interceptors to counter the attack missiles. The problem is computationally complex due to the presence of enormous state space. The Learning vector quantiser–Radial basis function (LVQ-RBF) multi-agent hybrid neural architecture is used as the learning structure, and Q-learning as the learning method. The LVQ-RBF multi-agent hybrid neural architecture overcomes the complex state space issue using the partitioning and weighted learning approach. The proposed LVQ-RBF multi- agent hybrid architecture improvises the learning performance by the local and global error criterion. The state space is explored with initial coarse partitioning by LVQ neural network. The fine partitioning of the state space is performed using the multi-agent RBF neural network. The discrete reward scheme is used for LVQ-RBF multi-agent hybrid neural architecture. It has a hierarchical architecture which enables quicker convergence without the loss of accuracy. The simulation of the TMD is performed with 500 assets and six priority of assets.
| || Connectionist Expert System to Diagnose Neck and Arm Pain
Author : Thamarai Selvi, S.
Source : Defence Science Journal ; Vol:49(3) ; 1999 ; pp 197-210
Subject : 612 Physiology;61 Medical Sciences;616.8 Neurology
Keywords : Neurophysiology ;Physiology;Medical sciences
Abstract : A connectionist expert system (CES) called BIONET aimed at assisting physicians in the diagnosis of diseases, such as neck and arm pain has been proposed. BIONET is an artificial network or connectionist network model capable of classifying diseases. Need for the development of CES for defence personnel has been discussed: BIONET is a feedforward three layer neural network with one hidden layer. The input, layer has been designated as stimulus layer, the hidden layer as receptor layer and output layer ag cortical layer. The sequential connections with spatial orientation have been maintained between stimulus layer and receptor layer for each specific factor. Parallel connections are established only at the cortical layer. Direct firing and facilitatory and inhibitory mechanisms are adhered to the neurophysiology of human nervous system. An algorithm for training on BIONET is also given. BIONET is simulated on a digital computer with training samples of patients collected from various hospitals in Tamil Nadu to diagnose neck and arm pain,diseases for testing purpose.
| || Solving Battle Management/Command Control and Communication Problem using Modified BIONET
Author : Thamarai Selvi, S.;Malmathanraj, R.
Source : Defence Science Journal ; Vol:56(4) ; 2006 ; pp 627-636
Subject : 621.38 Electronics;681.3 Computer Science
Keywords : Reinforcement learning;Modified BIONET;Radial basis function neural network;Fuzzy inference system;Multi-layer defence;Battle management
Abstract : This paper proposes and implements a neural architecture to solve the weapon allocation problem in the multi-layer defense scenario using modified BIONET neural network architecture. The presynaptic layer of the modified BIONET reduces the dimensionality of the principal state equation by partitioning the state space. The post-synaptic layer of the modified BIONET includes the perceptron Q-learning rule. The cortical layer incorporates L-learning scheme to provide better exploration over action space. Thus, action selection is effectively made with quicker convergence of training. The reward scheme in the reinforcement learning is obtained by calculating the measure of probability of survival. The decision module has been enhanced by incorporating the features corresponding to the battle weapons for effective representation of the environment. Thus, the modified BIONET neural architecture is used to increase the efficiency of assets saved in the simulation and the time complexity is reduced due to the state-space partitioning scheme involved in the neural network. The proposed modified BIONET is implemented in MATLAB and the percentage of assets saved is increased. Also, the training time is drastically reduced. Thus, the modified BIONET resulted in saving more assets with faster convergence of learning.