It is set of interconnected neurons which are used for universal approximation. Artificial neural networks are consisted of interconnecting artificial neurons (mimic the properties of biological neurons). Artificial neural networks can be either used to gain understanding of biological neural networks, or for solving artificial intelligence problems. The real biological nervous system is extremely complex: artificial neural network algorithms attempt to evaluate this complexity and focus on what may hypothetically matter from the information. Good performance or human error pertaining to pattern that can be used as one source of evidence towards the supposition that the abstraction really apprehend something important from the point of view of information processing in the brain.

Artificial neural network essentially formed with weights. Completely, when an artificial neural network is originally presented with a pattern it performs a random ‘guess’ as to what it might be. It then examines how far its result from the genuine one and makes a suitable adjustment to its attachment weights as shown in Figure.


Figure. Artificial neural network

Activation Function

An artificial neural network can be encouraged to execute a particular function by modifying the values of the weights among components. Artificial network function is calculated by the interconnections between components. This function is used to generate the relevant result from the weighted sum of inputs. The output is compared with the desired results; if the output generated is compatible with genuine output then the input is correct unless that output improves according to the weight.

Training of ANN

The need of ANN training of ANN is dependent on certain learning processes. Using learning system, a network is forced to generate a particular acknowledgment to a specific input. It becomes important when the information about input is unknown or inadequate.

Supervised Learning

In this scheme, we have assumed that at each moment of time when the input is utilized, the desired acknowledgment of the system is achieved. It works to predict outcomes for known instances. Such a system compares its predictions with the known results and learns from its mistakes.

Unsupervised learning

In this mode, the desired acknowledgment is not known, thus, explicitly error message cannot be used to enhance network behavior. As no information is provided as to correctness or incorrectness of responses, learning somehow is performed based on measurements of response to inputs.

Advantages of neural network

The various advantages of using neural network are given below:

  • Unlike rule based systems or programmed systems, neural networks are flexible in changing environment. Rule based systems are limited to the situation for which they have been designed. If the situation changes they are unable to operate in changed environment. Though neural may take time to learn a sudden change, they are good at adapting changing situations.
  • Since the system is developed through learning and not by programming, neural network teach itself the pattern. It can be implemented in any application without any problem.
  • Pattern Recognition is a powerful technique for harnessing the information in data and generalizing about it. Neural network learns how to recognize the patterns which exist in the data set.
  • Neural network build models that are more reflective of the structure of the data in significantly less time.
  • Neural networks can easily build informative models because they can handle very complex situations as they can easily handle data unlike the traditional methods like programming logic etc.
  • Neural networks operate well with modest computer hardware.
  • A neural network can continue without any problem even if an element of neural network fails.

Limitations of neural network:

  • Key limitation of neural network is its inability to explain how the network has been built.
  • Extraction of rules from neural network is difficult.
  • Time consuming process of training the neural network from complex data set.
  • Neural network needs training to operate.
  • Architecture of neural network is different from the architecture of microprocessors therefore needs to be emulated.
  • Requires high processing time for large networks


Leave a Reply

Your email address will not be published. Required fields are marked *