Research Review: Adapting and Enhancing Mussels Wandering Optimization Algorithm for Supervised-Training of Neural Networks

 

Research Review: Adapting and Enhancing Mussels Wandering Optimization Algorithm for Supervised-Training of Neural Networks
2014-03-11
Time 1500 until 1600
Meeting Room 7th Floor
Ahmed A. A. Abusnaina
Prof. Rosni Abdullah
N/A
Neural network (NN) uses mathematical models for information processing in order to accomplish a variety of tasks, which have been widely applied in engineering and sciences for diversity of applications. NNs are classified according to their computational units (neurons) as three generations: 1st:Networks based on McCulloch neuron, 2nd:Artificial Neural Networks (ANN) and 3rd:Spiking Neural Networks (SNN). Networks of spiking neurons have been considered more biologically realistic and more powerful than their non-spiking predecessors as they can encode temporal information in their signals. Also, SNN differs in information processing inside its basic computational units “neurons” and the network structure by adding delay synapses between its connections. The training process of NN deals with adjusting and altering the weights and/or structure of the network depending on a specific training algorithm. Training of NN fall into two main categories: traditional training algorithms and evolutionary-based training algorithms. Traditional training algorithms have several drawbacks such as local minima, training oscillation, and its slowness. Therefore, evolutionary-based training algorithms that depend on global optimization methods are utilized to train NN to overcome the drawbacks of traditional learning algorithms. New evolutionary-based methods for training NN are proposed. These methods are based on adapting and enhancing the Mussels Wandering Optimization (MWO) algorithm for supervised-training of both ANN and SNN.
N/A
none
2013/14
2
N/A