Artificial Neural Networks (ANN) are used in computing reasonable solutions to NP-hard problems. An artificial neuron has a number of input stimuli, and an output. Each neuron calculates an activation-value depending on the inputs and its internal weights and an output depending on the activation and a threshold. An ANN is a network of neurons where outputs of some neurons become inputs to others. ANN's are trained by presenting them with examples of a concept or by unsupervised learning. An ANN learns by adjusting its weights according a learning rule. They have been used to solve a variety of problems including, character recognition, detection of patterns, non-linear optimization, etc. This study explores ANN's and their applications to various problems. We have implemented several ANN topologies including: feed-forward, back-propagation, Kohonen, Hopfield, etc. Our implementation was successful in recognizing several English characters. We used the Hopfield-Tank model to implement a solution to the traveling-seller problem. We also compared this approach to other evolutionary techniques.
|Presenter:||Steven Klein (Undergraduate Student)|
|Time:||9 am (Session I)|