Artificial Neural Network

An artificial neural network (ANN) is a part of a computer system based on artificial neurons that mimic the behavior of human neurons. Each neuron processes various forms of input information and creates a single output unit depending on the activation function—that is, the weight and type of its connection with other neurons. Neural networks use a technique called backward propagation to recognize patterns in the information it processes and refine the weight of its internal connections to reduce errors.

Computer systems that use artificial neural networks require a huge amount of memory bandwidth, and limited bandwidth can cause bottlenecks that diminish the network’s performance. With adequate resources, however, neural computers are able to operate on abstract concepts and learn from the tasks they perform. In contrast, a traditional algorithmic computer is only able to execute functions for which it has a set of rigid, predefined rules.

Artificial neural network and neural network are often used interchangeably.

Types of artificial neural networks

There are several types of artificial neural networks with different structures that serve different purposes. New network structures are developed every day, so a comprehensive list of types is impossible. The most prominent types of networks and a diagram of their structures are detailed below.

• Feedforward neural networks are the simplest and most common type of network. It contains three segments (input, output, and hidden layers in between) through which information passes linearly.
• Recurrent neural networks look similar to feedforward networks, but with one major difference: the output is cycled back to the input layer to help predict the next output. These are frequently used for autocompletion functions.
  
• Radial basis networks use the distance of multiple points relative to the center as an activation function. They’re often used in power systems to prevent blackouts and restore power as quickly as possible. 
  
• Convolutional neural networks segment input data into multiple layers for sequential processing. Each layer is only connected to its neighboring layers, as opposed to all layers being interconnected. This type of network is commonly used for image recognition and natural language processing.
  

Source

How are artificial neural networks used?

Artificial neural networks are fundamental to artificial intelligence and deep learning. Each type of network usually has some combination of four main functions:

• Clustering, or arranging data according to shared properties
  
• Classifying, or sorting data according to predefined categories
  
• Predicting, or anticipating future events/trends based on data patterns
  
• Associating, or relating unfamiliar data to established patterns

These functions are useful across a number of different disciplines and industries. As such, neural networks have practical applications in financial trading, social network filtering, fraud detection, data mining, criminal justice, medical research and diagnoses, customer segmentation, and biometric analysis, among numerous other uses.

 

Related Links

• Why is so much memory needed for deep neural networks?
  
• Real-Life and Business Applications of Neural Networks