What is Neural Network?

Artificial Intelligence – A device takes in information , do some processing to complete the task successfully.

A system than perceive information from the environment , understand and interpret the data to take required action is known as Artificial intelligence machines.

A system which maximizes its chance of success by properly analyzing data is the core of Artificial intelligence. To develop Artificial intelligence (AI) products hardcoded instructions or program is of no use as the data around us is huge.

So as to develop a generic solution algorithms are designed in such a way that enables machine to learn the required pattern , gain intelligence overtime to take make good decision.

How Human Learn?

Acquiring new things and updating our skills , knowledge is the process of learning.God blessed humans and every creature on the planet with this skill. With experience we learn and grow.

Due to continuous interaction with society and environment  consciously and unconsciously we all are learning.

Five traditionally recognized senses of humans are: sight (eyes), smell (nose), touch (skin), hearing, taste (tongue). All the species in the world has multitude of senses.

With these senses we continuously gather information, neurons carry information our brain interprets the signals, processes, integrates and coordinates to take decision.

How AI mimics biological Neuron??

The brain basic working unit is neuron. Human brain is made up of 100 billion neurons. These neurons transmits information to and from the brain and to the various parts of the body.

Biological Neurons:

Deep Learning is a subset of Artificial Intelligence which consists of an algorithms inspired by the function and structure of the brain.

Artificial neurons are inspired by the biological neurons.The neurons are connected to each other to control body functions, emotions and movements.

The key components of biological neurons are: Dendrites bring information to soma so dentrites accept stimuli from an external environment.

Soma is the spherical part of neuron, the incoming signals are summed up by the soma when sufficient input is received neurons fires up(i.e. when threshold is exceeded) and axon sends information to other neuron depending upon the strength of the signal . If input is not sufficient no potential action is taken.

Artificial Neuron:

An elementary unit in Artificial Neural Network is Artificial Neuron. The artificial neuron is primarily composed of Inputs, weights, Activation function and Output. Each input has an associated weight (w).The input is summed up and non-linear activation function is applied to it. The output is given at the output line.

Working of Artificial Neural Network :

  1. Multiply inputs by its weight. For example : x1.w1j
  2. Calculate the weighted sum for each input and weight. ?wj
  3. The netput is given to activation function to determine the output. If the weighted sum is greater than threshold value assign 1 and else 0 as output.

Artificial Neural Network (ANN):

Artificial Neural network are composed of multiple nodes which takes input process them and give output. Each node output is known as activation or node value.

ANN consists of input layer, hidden layers and output layers. At hidden layer input is transformed to derive some pattern which can be given at the output layer.

Types of Neural Network in Artificial Intelligence: 

  1. Types of Neural Network:The Different types of Neural network are:
    1. Convolutional Neural Network
    2. Feed Forward Neural Network
    3. Radial basis Function Neural Network
    4. Multilayer Perceptron
    5. Recurrent Neural Network
    6. Long short-term memory
    7. Long short-term memory

    Let’s understand Convolutional Neural Network (CNN) with Example:

    Convolutional Neural Network works best for Images and videos which is a class of Deep Neural Networks. It has input layers, output layers and multiple hidden layers which are series of convolutional layers.

    The building blocks are

    1. Input Layer
    2. Convolutional Layers
    3. Pooling Layers
    4. Fully connected Layers

    As we can see in the diagram first input image with its height and weight parameter is feed at the input layer. The image is convoluted and passed to the next layer. Pooling layers is generally use to manage the dimensions of the data. Finally, the output is flattened and given at the output layer.

  2. Feed Forward Neural Network

It is one of the simplest form of the network where data (input) travels in only direction. In this network there is no back propagation and connections between nodes do not form a cycle or loops. Perception is the simplest feed forward neural network. 

  1. Recurrent Neural Network (RNN)

    It is a network where the flow of the data is not restricted to one direction. It has more capability and greater learning speed which is used to solve complex tasks.  In this network output from the next step is feedback to the previous state. RNN has memory which retains information for processing

  2. Bayesian Network:

The draw a probabilistic relationship between set of the variables Bayesian network is used.

They are used for wide variety of tasks such as decision making, prediction, anomaly detection etc.

It is a cyclic graph that denotes both random variables and conditional dependencies.

  1. Modular Neural Network:

It is a collection of neural network working independently where each neural network has a set of inputs.

The advantage of Modular Neural network is that it divides the large computational task into smaller modules and thus decreases the complexity.

Applications of Artificial Neural Network:

  • Hand writing Recognition
  • Face Detection and Recognition : In the upcoming days face Recognition will be a popular biometric. To extract features from millions of faces is a big task which can be accomplished by Artificial neural network.
  • Image Compression : In this era data is generated every second. Most of the applications and sites are using images either data transfer takes place or images are loaded on the websites so using neural network to reduce the size of image is worth.
  • Speech Recognition: To eliminate the communication barrier between humans and computer sophisticated neural networks can be made to understand spoken language of humans. Attempts to make speech recognition system are made using Multi-layer networks, Kohonen self-organizing maps etc.

Learn Deep Learning and Artificial Intelligence

In the similar way numerous applications can be made using Artificial Neural network.

Artificial Neural Network model is very good for problem-solving which are flexible and powerful. Without doing any hard-coded instruction just by training network with examples we can reap good results.

Deep Learning is the subset of Artificial Intelligence and it is a blooming field of this decade.