Artificial intelligence MACHINE LEARNING

              What is Machine Learning: A Conversation

Many people like us seem to talk about machine learning the future. Although nowadays it is appearing in every aspect of our life - whether it is Google's computer playing great go games or answering itself in Gmail's inbox. It all sounds very good to hear but still many people want to know what the machine learning is after, or why it matters, or why identifying a pet in the photo is not as easy as it seems. To understand this, we have talked with Maya Gupta, a research scientist in Machine Learning, Google.

        Let's start with the basic question. What is the machine learning after?

• Machine learning understands their patterns with some examples, and then uses that pattern to predict new examples in advance.

• Take the example of a movie. Let's say that one billion people tell about ten favorite movies. By using these examples, the computer can know what is liked in the movies that people like. Then the computer introduces these illustrative patterns, such as, "People who like horror movies usually do not like romance, but people like movies from one actor." Then if you tell the computer that you have a jack Nicholson's film 'The Shining' liked it, it could guess whether Jack Nicholson's 'Something's Gota Guy' would also like you or not, and who are you on YouTube? Suggest videos from.

         A little bit understood. But how does this actually work? 

• In fact, the patterns that the machine learns can be very complex and can be very difficult to explain in words. Like Google Photos, which lets you find dog photos from all your photos. How does Google do this? First we gather examples of photos labeled "dog" (internet thanks!). We also store photos with "Cat" and other millions of labeled photos, but I will not talk about them here :)

• Then with the help of the pattern of the pattern and color of the computer pixel, it estimates whether it is a cat or a dog (or something else). First of all, the computer predicts what patterns can be good for identifying dogs. Then it looks at the dog's photo used in the example to check whether its patterns are working properly. If it is accidentally understanding the cat as a dog, then it makes minor changes in the pattern used. Then it looks at the cat's photo and improves it again to get the correct pattern. And this process is repeated over and over again - look at an example and if it does not look correct, then improve the pattern used to improve that example.

• After this, machine-learning models are formed from patterns such as deep neural networks, which can identify dogs, cats, fire fighters and many such things correctly (mostly). 

This seems to be the future of the future. What other Google products do you currently use machine learning?

• Google is doing a lot of new things from machine learning, like Google Translate, taking a photo of the sign board or menus written in a language and locates words and language in it and translates it in your language right there. .

• You can speak about anything from Google Translate and the speech recognition that works through machine learning will mean that the technology of speech recognition will start. The technology that recognizes the speech is also used in the product, such as understanding your voice query in Google App that is the spoken question and making YouTube video more easily searchable. 

Is Machine Learning and Artificial Intelligence the same? 

Actually, these words can mean different things for different people, but Artificial Intelligence (AI) is roughly a word for computer programs that try to solve problems that people easily You can, like seeing a photo and tell about it One more thing that people easily do is learn from examples. And the machine learning program also tries to do the same: Tell the students to learn from examples.

The interesting thing is that when we understand how to create such a computer program, we can teach them to process a lot of data faster so that they can accomplish even hard work such as mastering the board (board game) Routing all people together in traffic, improving the use of energy throughout the country, as well as my favorite job - to find the best search results for you on Google. 

      Why is Google machine learning so important now?

• Machine learning is not new, but its roots can be found in 18th Century figures. But you are right that a lot of attention has been given recently, which has three reasons.

• First of all, we have to gather a lot of examples so that computers can learn to make better guesses. Then whether it is for those things, you or I can do it very easily (like finding a dog in a photo). Due to increasing activities on the Internet, we now have a large source of examples from which computers can learn. For example, now on the websites of the world, in every language, there are millions of photos labeled "dogs".

• But having too many examples is not enough. Now by showing some pictures of dogs to a webcam, they can not expect to learn anything from them - the computer needs a learning program. In recent years, some companies and people, including Google, have done such encouraging work that show how complex and powerful the machine learning program can be.

• Now our programs are not such that they can be fully trusted, as well as for computers, there is still a lot to learn, so to make a correct identification, if there is a slight change in the pattern, then we have many examples It has to be seen. All this requires a lot of computing power, and simultaneous processing. Although there has been a lot of development in the software and hardware sector, it is possible to do this. 

Any such thing that computers can not afford today, but with the help of machine learning will you be able to do so soon? 

• Until just a few days ago, the technology that recognizes the dialect was also difficult to identify ten number of your credit card on the phone. Due to machine learning in the last five years, the technology recognizing the bid has greatly developed, and now you can use it for Google search. And it's getting better, faster than this.

• I also believe that we can look better through machine learning. Do not know about you, but I do not like to wear them before buying clothes! Because of this, as soon as I get a brand of right fiting jeans, I buy her five jeans. But by learning from machine learning we can see examples of brands that are good for fitting, and which brands we can buy. Google is not working on this but I hope someone, somewhere, will be working on it!

How will the machine learning form in ten years?

• The one thing that everyone in this field is working on is how to learn faster than fewer instances. One way (on which Google is working very hard) is that we give our machines more common sense, which is called "regularization" in the field.

• What does the Common Sense mean to the machine? One meaning of this is that if some example changes a bit, then the machine does not completely change its decision. For example, keeping a cowboy hat wearing a dog's photo in a dog-labeled photo.

• In order to put common sense in the Learning program, we are teaching machine learning to ignore the small, non-critical changes such as cowboy hat. This thing sounds easy to say, but if you make a mistake in it, the machine will not pay any attention to the important changes too! That's why we are trying to balance the work.

BY: AMAN KUMAR SINGH