Neurons are brain cells. Artificial neural networks are engineering simulations -- in hardware or software -- that behave like interconnected neurons. Neural coding is the pattern of stored information in biological or artificial neural networks that allow them to give an appropriate response for a specific pattern of excitations. The neural code is developed by repeatedly exposing the neural net to excitation patterns and comparing their outputs to known correct outputs -- adjusting the neural code each time so that the response is a little more appropriate. If the training examples consist of time-related stimulus-response information, the neural net can learn to be predictive.
Train the neural network to set its neural code. After the neural code is set properly, the neural net will act as a predictor. When shown a pattern of inputs, it will predict the appropriate output.
Divide the training data into two random sets. Call the sets the "training" set and the "testing" set. Present each example from the training set to the neural net and check its response against the proper response. Use one of the updating algorithms to adjust the neural code to move the actual answer closer to the correct answer. Cycle through the examples in the training set several times.
Test the predictive accuracy of the neural coding by presenting the examples in the testing set to the neural net. Do not change the neural code during the testing phase. This phase determines how well this specific neural coding is working.
Verify that the trained neural net has the neural coding to accurately predict the phenomena that you trained it to predict. If it does not, you need to retrain it with new data -- perhaps a bigger or more diverse training set. If the net still does not behave correctly, you need new network architecture or a different training algorithm.
- Photo Credit Comstock Images/Comstock/Getty Images