What is machine learning to a kid?

 

As a kid, you must be excited to know about machine learning, as the name suggests it is a process of teaching machines to help them learn new things with the help of certain rules or by its own, exactly the way we humans learn new things.

The Definition

Let us first understand that machine learning is a process by which includes two things, the first part is a machine and the other part is learning. Now, we must focus on the word learning which is a process of identifying and classifying objects based on certain features and characteristics. 

How it works

We can teach machine certain rules and systems of classification. Based on the rules, the machine learns to identify and classify objects based on certain characteristics which it has learned. 

How the machines are trained

Training of the machines is done with the help of suitable examples over a period of time. This particular process of training the machine in order to clearly identify and act according to the requirement of the user can be defined as machine learning. For example, if you want your computer to identify your friends by name what you need to do is that at first, you need to train your computer with the pictures of your friend and the characteristics of each of your friend along with their name. 

 Why we need more examples

 The computer will learn with the examples that you have set, and then next time a friend comes to your home, the computer can identify and classify your friend and successfully name them. As more examples are fed into the system, more accurate classification and identification of your friends become easier for your computer. This particular process of helping the computer learn is defined as machine learning.

 The Next Step

 Now, you may be curious to know about the different types of machine learning which we will discuss subsequently. If you are interested to know more about machine learning, you can mail at smartsubu2020@gmail.com.

Read More

What are the different types of machine learning?

We hope you have understood about machine learning and if you have not gone through the post on machine learning, kindly  

As the name goes supervised learning is just like the learning process used in schools, where there are a set of instructions given by our teacher and we are asked to perform a certain specified task. But, in case of machine learning the training data set is provided to the computers, and the user acts as a teacher where the algorithms are designed  to give a very specific output (just as the students).

The machines are trained on the predefined data set before it can be used to make any decisions based on new data. For example, the machines are set to identify certain objects based on very specific characteristics which are very unique to that particular object.

Suppose a machine is trained on identifying whether a particular creature is a dog or cat. Specific features of dog and cat are fed to the computer during the training process and levelling of the pictures are done so that whenever new characteristics or features are fed into the computer it is able to correctly identify those features and come up with the outcome of whether it is a dog or cat.

The second type of learning is unsupervised learning where the machines are given a particular data set and left to learn on its own. To simplify, the machine is fed with data and the algorithms by itself tries to find pattern and features in the data set and put them into different categories by identifying patterns and correlation within the units of the data set.

It can be mostly said as a self-study process, where the machine learning algorithm is put into its own to identify and act by identifying relationship among the features of the different clusters. Clustering is one of the most commonly used unsupervised machine-learning techniques.

The Third category of machine learning is reinforcement learning, where the machines are left to learn by itself, but, here the algorithms of learning are rewarded or penalized based on the actions that they perform. For example as a child if you do good things you are rewarded, and hence, you are encouraged to perform that action again and again, whereas if you are given punishment for a particular act of yours you tend to not repeat it.

Similarly in reinforcement learning the algorithm is rewarded when it performs the correct task and thus the algorithm learns and tries to repeat the same task again and again. Whereas if penalized for a task it tries not to repeat those task.

In this reinforcement learning process, the machine learns by the trial and error method. In terms of machine learning that rewards and penalties are decided based on the predefined actions which the machine needs to perform in interaction with its environment.

If you are interested to know more about machine learning, you can mail at smartsubu2020@gmail.com.

Read More

What is the difference between machine learning and deep learning?

Machine learning and deep learning are more or less on the same genre of machine learning algorithms but with a slight difference in the way they are being executed.

In the case of machine learning, what we do is that first of all we try to identify the defining features from the data set which correctly defines or classifies the data and this process are known as feature extraction. 

What we do next is that we feed these features into our classification algorithms to train the machines in order to identify and predict the correct output.

Deep learning is a subset of machine learning where the feature extraction and classification is done simultaneously by the algorithm itself with the help of neural networks set for the particular purpose in the algorithm. 

Deep learning is more influenced by the working of the human brains where the inputs are fed and the processing is done in the hidden neural layers and finally, the outputs are produced. 

Deep learning algorithms have become more popular in case of image recognition, computer vision and natural language processing.

Artificial Neural Networks, Recurrent Neural Network and Convolution Neural Network have played a significant role in deep learning. Don’t worry if you don't know anything about these neural networks we will discuss this particular ANN, CNN and RNN separately.

Read More

How can you differentiate between regression and classification?

Regression and classification both belong to supervised learning. Regression is mostly based on the data which are continuous in nature whereas the classification algorithm is mostly based on the labels provided to the data set. 

Regression is based on the paradigm of continuous prediction whereas classification is mostly used for predicting the probability of a particular object belonging to a class.

The outcome in the case of classification problem are mostly binary in nature, for example, whether the customer will buy a particular product or not, whether the particular student will pass or not, whether the particular interview will be cleared or not and many more.

Regression algorithms are mostly used for prediction of a continuous variable in the future time period based on certain inputs which can be continuous or categorical in nature.

In the case of classification algorithm the inputs and outputs are both categorical in nature and the probability of identifying a particular class is already defined in the classification problem. For example, the number of classes which will be produced in the classification algorithm is already predefined in the algorithm by the user.

If you are interested to know more about Regression and Classification, you can mail to smartsubu2020@gmail.com.

Read More

What is selection bias?

In order to understand selection bias we must understand how particular machine learning algorithm works. In order to train the machine learning algorithm, we need a lot of data. If we can comprehensively collect all the data and then feed that into the machine learning algorithm there is nothing like that.

But, it is not always possible due to time and money constraint, hence we resort to sampling.

Sampling is a process of collecting a small proportion of the data set which we assume to have all the characteristics of the population we are interested to study. In the process of sampling certain discrepancies may creep in while selecting a particular data set from the whole and this is known as selection bias.

To elaborate simply, if you are asked to select 100 people from your town which has a population of about 1 lakh it is most likely that you will be recalling the names of people whom you know. This particular type of bias is known as selection bias and it often dilutes the characteristics of the training data set which is used to train the machine-learning algorithm and hence the learning process is compromised.

More robust the training data set, more is the probability of the machine learning algorithm to predict correct results with new data sets.

If you are interested to know more about Machine Learning, You can mail to smartsubu2020@gmail.com.

Read More

What is true positive, false positive, false negative and true negative?

In machine learning in order to judge the performance of the algorithm Precision and Recall are used. We will discuss Precision and recall later, but before that, we need to understand the four things which are very important in precision and recall.

These are true positive, false positive, false negative and true negative. We will start the discussion with the help of an example, suppose we have designed an algorithm which is supposed to predict correctly the event of a fire. So, when there will be fire and an alert alarm will set on. We will take four scenarios one by one.

Scenario 1:  there is a fire and the alarm goes on in this case the algorithm is correctly able to identify the event and the action is also correct hence this is known as true positive. Another simplest way of understanding this is that the actual event has occurred and the prediction of the algorithm is also correct hence it is true positive.

Scenario 2: there is no fire and the alarm also does not ring. In this case, the event has not occurred and the algorithm is also able to correctly respond to the event and the action is also correct. The actual event has not occurred and the prediction of the algorithm is also correct in predicting that the event has not occurred hence it is true negative.

Scenario 3: there is no fire but the alarm rings. In this case, that event has not occurred but the algorithm has predicted that the event has occurred, hence it is a false positive. In simple terms here the algorithm has misclassified the event and hence it is a false positive. 

Scenario 4: there is fire and the alarm does not ring. In this case, the event has occurred but the algorithm has not predicted it correctly. In simple terms, the event has actually occurred but the machine was unable to identify it correctly hence it is known as a false negative.

To sum up, the classification power of the algorithm depends on the amount of true classification the algorithm has done in terms of the number of true positive and true negative. The performance of the model is considered to be good.

Whereas misclassification happens when the number of false-positive and false negative is more and in this case, the model did not perform well.

In the next discussion will be taking up the issues of Precision and recall in machine learning.

If You are interested to know more about Machine Learning Algorithms, you can mail to smartsubu2020@gmail.com.

Read More

What is Precision and Recall in machine learning?

In order to understand Precision and Recall, we will take a very simple example. Suppose we are asked to recall certain events from past. For the sake of simplicity, let us take that there are 20 very significant events in our life which we're asked to recall.

If we are able to recall 20 out of 20 events then our recall ratio will be 1 or 100%, but if you are able to correctly recall only 10 events out of 20 events then recall ratio will be 0.5 or 50%.

Whereas, precision means how many events we can correctly recall out of a total number of events which we recall (sum of both correct and incorrect events).

For example, if we tell that ok, I have correctly recalled 15 events but actually, I have only correctly recalled 10 events, in this case out of 15 times of I am precise in 10 events.

So, precision can be defined as the ratio of the number of events the algorithm can correctly recall out of the total number of events that which the algorithm recalls (sum of the correct events as well as the incorrect events).

In this case, we have recalled 15 events, but out of that only 10 events were correct, hence your precision rate is 10 by 15, i.e  66.67%.

Similarly, in the case of machine learning Recall and Precision are used to identify the performance of machine learning algorithms. Apart from that, there are other sophisticated measures like Sensitivity, Specificity, ROC and AUC which are also used for performance measurement of machine learning algorithms.

If this piece interest you, you can mail to smartsubu2020 to know more.

Read More

What is confusion matrix?

Don’t get confused with the word confusion matrix, there is nothing to be so confused about it. In simple terms, the confusion matrix is used to identify the best performing algorithms within several competing algorithms.

Please refer to the figure to have a view of the confusion matrix, it consists of true positive, false positive, false negative and true negative. We have already discussed about  true positive, false positive, false negative and true negative in our earlier sections.

If the number of true positive and true negative is more in the confusion matrix then the algorithm is performing well, but, if, number of false positive and false negative increases, the performance of the algorithm is not good.

The confusion matrix can have many dimension and it depends on the number of outcomes we want to predict, for example, if we want to predict for ‘number of outcomes then the confusion matrix will have n dimensions.

The diagonal elements in the confusion matrix will be those elements which are correctly classified by the machine learning algorithm.

Don't get confused about confusion matrix, it is just a matrix of identifying the best machine learning algorithm amongst competing algorithms.

If you are interested to know more about the confusion matrix, you can mail to smartsubu2020@gmail.com.

Read More

What is Infrastructure as a Service (IAAS) and Platform as a Service (PAAS)

Today we will be discussing about the deployment of models and about Infrastructure as a Service (IAAS) and Platform as a Service (PAAS) in the cloud networks. Before going into the details of the discussion, we need to understand what is needed for deployment of the model and what are the basic features which need to be taken care of in order to eventually go into the deployment models.

We need to first understand how we can create web services and then host the web services in the cloud. So, there is a need to understand what we mean by a server and the different type of infrastructure used in the web services in the server by using the server.

The service on which the applications are made to run can be either made at the local level, for example, in the office or any the data centre of the office to host the services. In the local level all the requirements of running a web service, for example, the application path, and the data requirement runtime needed, the middleware requirement, the operating system and the requirement of the service are taken care at the local level.

Amazon Web Services are popularly known as AWS provide IAAS. The requirements of the servers, networking and the storage is provided by the web services company and the application along with the data requirement, operating system and the runtime is decided by the company who is using the web services of the company providing the IAAS.

In another type of service all the requirements of the operating system, the run time, the service requirement in networking to be provided and the storage requirements are taken care of by the company providing PAAS.

The application path and the data required to run that application is decided by the creator of the machine learning models and hosted in the PAAS. In the platform as a service paradigm, it is very easy to deploy any machine learning model as one need to take care of only the application path and the data requirement path and leave everything to the platform as a service provider to put their model up and running in the web.

The costing depends on the scalability factor of the models and in order to understand the deployment of the application and the data requirement we need to understand the different dimensions involved in the process of deployment.

The most common framework used for creating the web service is a flask and it is python based programming which needs to be understood in order to create the web services and for the deployment of the models on the web.

Read More