Neural network software uses pattern recognition to predict a trend or identify an irregularity in behavior. While all computerized neural networks operate on the same general principle, this type of software can be used in many different ways. Three of the most common varieties are image, data, and voice neural network software.
All neural network software is based on the structure of a human or animal brain. As with a brain, the individual processing areas of an artificial neural network can be re-arranged into new structures. Instead of being coded to perform only one limited job, neural network software can be taught a general pattern and then use this model to predict the outcome of many different events.
Image-based neural network software is well suited for comparing pictures and processing visual patterns. This variety of program is usually linked to a high-resolution camera or a collection of existing images. In one example of image neural networking, researchers have used computers to visually categorize flowers into the correct plant species. Visual neural network programs can analyze attributes such as the length and color of an object, and sort images intelligently.
Some types of neural network programs are able to perform similar tasks in real time. Software can be connected to a surveillance camera, and observe an area for movement or behavior that is out of the ordinary. Some police departments and security groups use this software to reduce manpower requirements while monitoring an area virtually.
Other types of neural network software are designed to work with raw data. Numbers or text variables are usually provided to the network, which can process the data to find trends. The banking industry often uses this type of software to estimate the chances of bankruptcy or credit default based on available financial records. These types of networks can also be used to determine the value of real estate based on many different factors, or deduce the value of a company.
Neural networks can also be designed to process voice information. As with imaging processing programs, this type of software can compare two audio samples or explore the trends of many different voice records. Some voice recognition programs use neural networks to determine which word is being spoken. This can be used for automatic dictation, or for audio-commanded applications such as voicemail. The flexible learning capability of a neural network allows software to recognize words even if the speaker has an accent.
