I wrote an article “Neural Network Training using Particle Swarm Optimization” which appears in the December 2013 issue of Visual Studio Magazine (VSM). See http://visualstudiomagazine.com/articles/2013/12/01/neural-network-training-using-particle-swarm-optimization.aspx. A neural network (NN) can be used to make predictions such as predicting the likelihood of a surgery patient’s death (low, medium, high) based on factors such as age, blood test results, and so on. Training a neural network involves finding a set of weight values so that computed outputs best match known outputs for a set of training data. Once these best weight values have been found they can be used to make predictions on new input data where the result is not yet known.
By far the most common technique to train neural networks is to use what is called the back-propagation algorithm. Although mathematically elegant and relatively fast, back-propagation is extremely sensitive to parameters to control exactly how the algorithm searches for the best neural network weights. A rarely used alternative NN training technique, which is much less sensitive to initial values of algorithm free parameters, is particle swarm optimization (PSO). PSO is loosely based on coordinated group behavior such as that exhibited by flocks of birds. In the VSM article, I show exactly how to use PSO to train a neural network.