| Tip |
| One of the most important features of the ANNs is their immunity to noise. An ANN is able to learn from clean or noisy data sets. Once the network has been trained, it may be able to recognize specific patterns, or eliminate noise from these patterns. |
Auto-association |
| An ANN can be set up for auto-association by applying specific information at its input while expecting the same information at its output. In this configuration, the number of inputs and outputs of the ANN must be the same. Additionally, scaling information at the input and at the output must be equal. In the figure shown below, a segment of a sine wave is applied to the input of an ANN; the target at the output of the ANN is the same signal. Typically, the training set may be built by showing different segments of the wave to the ANN. The input may represent: an image, a sound, a signal, a vibration, an event, etc. |
| Problem 1 |
| List five applications where an ANN in auto-association configuration can be used. |
| Tip |
| In many auto-association problems, the number of inputs is large. Because the number of inputs and outputs must be equal, the number of outputs is also large. For these problems, the ANN may have many weights to adjust, and the required number of training case is typically very large. Thus, the training of an ANN for auto-association may be very time consuming. |