Q 359. What is a Neural Network? What are some of its usages in decision making?

 

Note for website visitors - Two questions are asked every week on this platform. One on Tuesday and the other on Friday.

• All questions so far can be seen here - https://www.benchmarksixsigma.com/forum/lean-six-sigma-business-excellence-questions/
• Please visit the forum home page at https://www.benchmarksixsigma.com/forum/ to respond to the latest question open till the next Tuesday/ Friday evening 5 PM as per Indian Standard Time. Questions launched on Tuesdays are open till Friday and questions launched on Friday are open till Tuesday. 
• The best answer is always shown at the top among responses and the author finds honorable mention in our Business Excellence dictionary at https://www.benchmarksixsigma.com/forum/business-excellence-dictionary-glossary/ along with the related term

Neural network, which mimics the way the human brain operates, is a circuit of neurons, which is also known as artificial neural network, that comprises of artificial neurons.
Neural network is used to solve problems in artificial intelligence (AI).
The way biological neural systems processes data has inspired how neural networks processes information.

Some of the uses of Neural Networks are:
->  video or online games
-> sequence recognition (handwritten text recognition, speech, gesture), 
-> decision making (virtual chess), 
->  diagnosis in medical field,
-> control (process, vehicle), 
-> pattern recognition (face or object ).
 

 

 

 

 

Neural networks are a mock of human brain where pattern understanding for given data is done via layers of neurons with each layer having specific number of neurons. Every neuron houses a function to add it's bit to overall decision or prediction of the network. Its applications can ben seen in predicting outcome of the process based on the input data derived from the same process. 

Neural Network:- Before understanding Neural Network, its imperative to understand a term called "Perceptron". Perceptron forms basis for Neural Network and also for Artificial Intelligence, hence the name has been given to one of the early algorithm.

Neural Network is a "Multi level Perceptron"  which works as  artificial "Neurons" of human brain and are arranged in a layered sequence to predict the patterns in the given data to get the output (Neurons are a micro particles in human brain which transmits information between different areas of brain by using electrical & chemical impulses).

 

To explain it in a simple and common terms, Lets say if we see a person our brain will calculate an approximate age of the person based on certain facial features like hair color, skin wrinkles etc.

Similar to our brain functioning, Neural network also will look for various patterns available in the data and will produce an output.

 

A Neural network is a two way data transmission methodology that it will check the output with the actual results and well move back to the different data patterns till the output matches with the actual results. 

 

Usage of Neural network:-

• Facial Recognition
• Business analytics
• Demand forecasting
• Trading systems

Neural Network:

 

Neural network from the base of deep learning and the sub-field of machine learning, where the algorithm is inspired by the structure of the human brain. Neural Network taken data trained themselves to recognize the pattern in the data and then predict the output for new set of similar data.

 

The structure of Neural Network is known as a multilayered perceptron, that is, a network of many neurons. In each layer, all the artificial neuron has its own weighted inputs, transfer function, and output. Once the Neural Network is trained and tested with the right weights decided, it can be given to predict the output. It gives high value during the decision making.

A Neural Network is sequence of mathematical and logical algorithms that work in tandem to identify the core link or connection in a set of data (input) through a process that follows the working of the human brain.

We know from various researches on how the various “Sensory” or “receptor” centers in the human brain are connected through a network of neurons that react to any stimuli/information from the “5 Senses”. These neurons then link through a network to draw up various permutations and combinations that can be derived from the information, based on the experience or training that these receptors have derived from the live world. The network follows a process or set of “learned” rules for calculation or for any other problem solving Operation. This is also known as algorithm.

Similarly as an analogy, a neural network is made up of many perceptron neurons, in layers, based on the “training” received. The said layers are also called hidden layers which are the primary unit that works together to form the Perceptron layer. These neurons are the ones which receive information or various forms of data, in sets of inputs. These inputs are combined with a bias and a group of weights which produces a single output, which can be a built up / calculated perception.

For this process of computation, each neuron considers assigned weights and bias. Then the “defined “combinations functions, across the network, uses the weight and the bias to give the output (perception) through the following equation:

Combination = bias + Wights * inputs

Post this, the activation functions produces the final output with the following equation:

Output= Activation(combination)

Thus the process flow can be briefly put forth as ;

1)    Information/ data is fed into the input layer and then transferred to the hidden layer(s)

2)    The interconnection between the said 2 layers assign weights to each input randomly

3)    A bias added to each input after the weights are multiplied with them individually

4)    The calculated sum is transferred to the activation function

5)    Which nodes it should fire for feature extraction is determined by the activation function.

6)    An application function is applied by the model to the output layer, to provide the output.

7)    Weights are adjusted and the output is back propagated to minimize the error.

Thus the final output is close to accurate if not accurate. It all depends on the incorporated “training”.

The accuracy of the output (perception) can be improved by

-      Increasing Hidden layers

-      Change the logic of activation function

-      Change the logic of activation function in the output layer

-      Increase the number of neurons

-      Supply optimal initial weights while training neural networks

-      Provide more of relevant data/input

-      Normalizing /Scaling the input data

-      Revise learning algorithm parameters

-      Use Deep learning / architecture for auto feature generation to enable every layer to refine the features.

-      Choose the neural network model according to the problem

Thus, Neural network models provide inductive means for collecting, storing and using experiential and realistic knowledge. Once the confidence on the accuracy of the output is arrived at, major business decisions can be taken based on the said outputs from the neuron network model. In short, decision support systems can depend on the output from efficient neural network models. We see in recent times that modular neural networks can perform highly complex task with great efficiency with very quick results.

Many decisions on sales forecasting, data validation, customer research, risk management and priority settings can be taken by business leaders, based on the outputs from modular Neural networks. In these times of penchant for disruptive decisions by entrepreneurs and investors, the Neural Network Models have gained a lot of prevalence!!

Rajesh has explained in detail the concept of Neural Networks and hence his answer has been selected as the best!