MACHINE LEARNING A PROBABILISTIC PERSPECTIVE:
MACHINE LEARNING A PROBABILISTIC PERSPECTIVE: Machine learning is a method of inputting something into a computer through a program. In these topics, we will cover What is Machine Learning, Machine Learning Algorithms, Machine Learning Vs Deep Learning.
Machine learning is a method of inputting something into a computer through a program. Analyze those things by looking at them, and giving a result based on those things that are currently being looked at. In this case, the programmer must also work on certain issues.
In his speech, he said that 8% of the 10,000 computer engineering graduates in the country are contributing to software development every year. The biggest risk in building a technology-based country is to create skilled human resources. As the number of students in computer engineering increases, they need to be encouraged to develop software.
The ‘Tensor Flow Developer Summit 2018’ conference was attended by machine learning developers of the country, officials of the Information Technology Department, and officials of telecom companies through which the progress and future prospects of machine learning were highlighted.
Arif Nizami, the advisor to Google Developer Group and chief executive of Preneur Lab, said the global market for artificial intelligence and device learning software continues to grow. According to a Google survey, the market for machine learning products and services will reach 8.3 trillion by 2035.
Rakhshanda Rukham, manager of Google Developer Group, said machine learning is the technology of the future. We want the country’s startups, developer companies to be ready in advance. Such an arrangement is for them.
The Google Brain team developed the first Tensor Flow for Google’s research and production work. It is currently the most widely used machine learning tool in the world. The use of tensor floor is behind all Google products.
A major breakthrough in machine learning would be equal to ten Microsoft. It’s not about me, it’s about Bill Gates himself.
What is Machine Learning?
Machine learning is a great subject in computer science. Usually we give some instructions to the computer, the computer works accordingly. But in the case of machine learning we tell some processes, the rest he learns on his own and works accordingly.
As for the weather, we will give all the previous weather data to a computer program. Our program will forecast the weather by analyzing the data. We are not saying here what to say. The program looks at the current weather by analyzing the previous data and tells us what the weather will be like tomorrow. This is machine learning.
Machine Learning Algorithms:
If you give data to a computer program, it will not be able to analyze it. We have to give some instructions on how to do data analysis. What machine learning algorithms to use, etc. If you do all this, the rest of the work will be done by the machine or the program itself.
Now even small apps need machine learning. Machine learning is applied in many areas of computer science and artificial intelligence, including data mining, natural language processing, image recognition, and expert systems.
Many examples can be given. Let me give you a close-up example. There is an Android app called Madviser. Its job is to suggest which mobile package would be better for you. And for that it analyzes all your data packages and voice calls.
This app is not told directly what to do. Giving some data, analyzing that data and suggesting a package to the user. It is also an application of machine learning. The app is made by Bangladeshi developers.
Another simple example is OCR or Optical Character Recognition / Reader. Used to read text from pictures. Many of us may have used it. This is also an application of machine learning.
Looks at the characters from the image and then detects what is what. And this machine is used in learning Classification Algorithm. Machine learning or classification algorithms are also used to find out which emails are spam and which are not. The use of many such algorithms is machine learning.
Machine learning algorithms can be divided into four main categories.
Supervised Learning: The program is trained on some pre-defined datasets. Gives program decisions based on train data. This is supervised learning. Whether the mail is spam or not, this decision is based on some previous data. This is an example of supervised learning.
Unsupervised learning: Unsupervised learning provides some data to the program. The program makes decisions based on that data. There is such a basket of fruit. The program will divide different results into different categories, this is an example of unsupervised learning.
Semi-Supervised learning: Semi-supervised learning is a combination of supervised and unsupervised.
Reinforcement learning: How do we learn in childhood? If I feel good after doing something, I do it more. If I get hurt by doing something again, I don’t do it anymore.
In fact, the way a person or an animal learns is exactly the way it is trained in a program in reinforcement learning. Siri, Cortana is the successful applications of machine learning.
After uploading the image to Facebook, it automatically detects the car’s image. This is also the application of machine learning. Bill Gates did not lie. If machine learning can be applied more efficiently, our whole technology world will change. Not that not being tried. This sector is being developed regularly.
There are two types of people. Watcher & Player. What kind would you be? If you are a player, you can study these from now on. You can know slowly. There are many resources available on the Internet. These two courses on machine learning are both great and popular.
ML Course at Udacity
ML Course at Coursera
Do a little search on Google and you will find all the great resources. ML Expert has a lot of demand. And there is a shortage in our country. Where there is a lack, there is an opportunity. You want to use it
Machine Learning Projects:
In machine learning you can use Linear Regression model to solve this problem. This allows you to estimate a number in a continuous range based on different independent variables.
Now see this is a straight line equation
I hope everyone knows this. Now xx is an independent variable and depends on the value of yy. This is a simple regression, which indicates a straight line equation.
The job of our model will be to determine the exact values of w1w1 and w2w2 so that they are able to make accurate estimates. On the other hand, in the case of multiplex regression, we are interested in determining the value of yy using multiple variables like xx.
In that case, however, the model would be the equation of a plane in a three-dimensional country and a hyperplane in a multidimensional country, instead of a straight line. Then we need to determine multiple constants like w1w1 and w2w2 to find the right hyperplane.
Now in our case
So all we have to do is determine the values of w1, w2, w3w1, w2, w3 so that the cost of our cost function is minimal. Our cost function is
We have discussed the cost function in the previous post. Here yiyi is the actual value and at money_to_spend we set the value estimated by the model. Maybe you’ve given a mm bar lover’s birthday gift before. You know the values of the then independent variables, but also the values of the subordinate variables.
This is how you determine the cost. Now we will reduce the value of this cost function as much as possible by using an algorithm called Gradient Descent.
Gradient Descent is a Parameter Learning technique that we can use to determine the value of ww or constants and reduce the cost. Now let us see a plot of Cost function with respect to any one constant
Here, for example, in Ww, the Slope or Gradient of the Cost function is positive. To get to the lowest part of our cost, we need to reduce the value of ww here. See again
Slope or Gradient of Cost function in ww is negative here. If we want to go to the lowest part of the cost, we have to increase the value of ww. But at the lowest part of the function, the slope is zero. We can then stop our parameter learning.
In this way we will determine the desired value for our constants. So we write this process mathematically in this way.
Here w is the special constant, gradwgradw is the Gradient value of Cost relative to that constant and alpha is a special parameter which we will call Learning Rate. This is a Hyperparameter that determines exactly how much we increase or decrease the value of our constant.
If the value of this alpha is too large, the value of our constant will come and go again and again; We will not be able to reach the minimum at the end. Again if the value of alpha is taken too small, the value of the constant will decrease or increase very slowly.
Note that as we get closer to the minimum, the value of the slope will automatically decrease as the angle between the tangent and the X axis decreases. So the parameter will also start to learn slowly.
On the other hand, the slope will continue to increase away from the minimum. And the parameter will try to learn in big steps. So if we use big alpha, we will never reach the minimum.
For our model
This way we can apply Gradient Descent.
Machine Learning vs Deep Learning:
You May Also Read: