Artificial Neural Networks and Machine Learning

Let's take a look on Artificial Neural Networks and Machine Learning. I know you’ve been started thinking about neuron as a biological term (that is in our body for conducting the reflex actions) but the same concept of neuron is used to create deep learning and machine learning models. They are called as Artificial Neural Networks. Basically, Neural Networks is used to create an artificial memory stimulation like humans because we humans learn by observations, experiences and practicing. 

ANN is defined by layers basically, it has input layers, hidden layers and the output layers. From input layer, user is required to give the data for training purpose, and after that these data sets get into the hidden layers (there can be many hidden layers and all are interconnected with all the nodes of input layers and next layer’s node respectively) these hidden are made using advanced scientific mathematical computations and these are available inside any basic machine learning library. And after calculation from the hidden layers’ machine gives the output data through the output layer (each node of output layer are connected to every node of its previous hidden layer). For training the models we can use ML libraries like scikit-learn, keras, tensor-flow, etc.

  

                        

Basically, what happens here is like, we can understand it from example that we want to detect a number 2 from a given input image then input layer takes image as input and the hidden layer processes the image and breaks the image into pixel and then analyses all the pixels using the layers and if there is a dark pixel then there is probability to get the digit piece so more is the dark pixel means more is probability nearer to 1 and if it is blank or white pixel then probability tends to zero and using this detection algorithm machine detects the number.

                       

This event of processing data and giving output in straight forward manner is known as forward-feed or Forward propagation.
But after processing the data and giving output machine learns from the output data by fitting it the curve of the previously labelled data and if there are variations found, they result in error and using machine learning algorithm the machine learns from it and make sure that next time when the processing happens it minimizes that error and it learns again and again to make the output data in order to achieve the accuracy and get precised data and this process of learning from the previous data by going by layers in reversed order is called Back-feeding or Back Propagation.

In the similar way, humans want to give computers the ability to think and recalculate by learning from previous mistakes which is the basic of machine learning (ML). In Machine Learning, want our created model to predict based on different learnings.

Types of Machine Learning are:

•  Supervised learning.
•  Unsupervised Learning.
•  Reinforcement Learning.

Supervised learning is the Learning is the learning based on the given labelled data. Labelled data is a data in a structured format and the question as well as the answer contained within it. We can think of a straight line equation to understand this in better way line has equation y =mx + c. in which f(x) = y and for each value of x we get a corresponding f(x) value.

     

So basically labelled data is the data of x and f(x) taken by many observations and when it is given to input layer ANN for learning then all the inputs are used for training the machine using machine learning libraries which use the complex algorithms and everything is basically based on the mathematical computations in background. So to learn the machine learning you need to have good foundation in the subject of mathematics.

For supervised there are so many Algorithms, out of which Linear regression is the basic one. We can simply understand the concept of linear regression by relating it to the straight line equation [yi] = ai[xi] + bi.

In this case, machine uses the basic algorithm to predict the upcoming values like we can predict whether the x and y satisfies the satisfies the, calculated equation then it will exhibit the same behavior (graph) and hence machine predicts the
value.

There are many types of learning methods like linear regression, random forest, k-means clustering, support vector machines.
All the algorithms are designed for special type of learning like for classification, for object detection, etc.

On the other hand, the unsupervised learning and the reinforcement learning use the unlabeled data and uses advanced learning algorithms for training the datasets.

Most of the times python are recommended for machine learning because python offers a huge library modules related to the machine learning, deep learning, and all other fields than all other languages.