Artificial Intelligence: Greate revolution in Technology is a great sentence with great knowledge.
What is Artificial intelligence?
It a very large branch of Computer science in which we make the computer enough smart to solve some real task as a Human intelligent mind does. Under AI, we use Machine Learning and deep learning in every sector of the tech industry. From Facebook to YouTube everyone is using AI to give users a great experience.
When we talk about AI, the first thing people think of is robots. This is not their fault because what we see from childhood, we think that is right and we adopt that knowledge. We see so many robots as look like Humans in many of the films that are oriented on technology and conclude that AI is only to have robots that can do the task of humans as we see in the film “ROBOT 2.0 – Chitti robot”.
Well, this is true that robots use the AI but you have only half knowledge. Actually, AI is just only the concept under which we use Machine learning, Deep learning, and neural network, and many more.
“Artificial intelligence is a set of algorithms and intelligence to try to mimic human intelligence. Machine learning is one of them, and deep learning is one of those machine learning techniques.”
Two Types of Artificial Intelligence are –
- Narrow AI: These types of AI are used for limited work and operates within a limited context and is a simulation of human intelligence. It is all around us and we use daily for some purpose.
A few examples of Narrow AI include:
- Google search
- Image recognition software
- Alexa, Siri, Cortana, Google, and other personal assistants.
- Self-driving cars
- Artificial General Intelligence (AGI): It is also called Strong AI. It is used while creating real robots just like you have seen in the films. These robots are so intelligent that can solve any type of problem you give them.
AI uses machine learning a lot to find the insight data and also to predict the data that is not available by using the historical dataset.
Artificial Intelligence applies machine learning, deep learning and other techniques to solve actual problems.
What is Machine Learning in Artificial intelligence?
Machine learning is one part of Artificial Intelligence that provides a system to learn the pattern and then apply it to the bigdata and find the particular things.
Machine Learning has algorithms that can be used to learn the data pattern by the computer without the help of any human coaching or guidance. These patterns are so useful and can be used to predict future data.
“More and more data you provide, more and more precisely it predicts the future data”
Machine learning algorithm are divided in three categories –
- Supervised learning algorithm – This learning algorithm is given the name or label of data and the desired output.
- Unsupervised learning algorithm – This learning algorithm is given no names or labels and is asked to identify the pattern in the input data. Uses – it is used in the recommendation system in an e-commerce website.
- Reinforcement learning algorithm – This learning algorithm is used in self-driving cars where the environment is dynamic.
Why we use Machine Learning?
- Finding and summarizing relevant data
- Making future predictions based on data analysis.
- Finding probabilities for the specific results and many more.
Use case: Why You tube use Machine learning in their website and App?
With 5 billion-plus downloading from the play store and more than 1.9 billion users logged to YouTube every single month. With having watch time over a billion-hour everyday YouTube has become the largest multimedia platform. On an average every minute, creators upload approx. 300 hours of video on this platform.
I don’t know have you ever wonder or not that how youtube know your choices and recommend only those videos which you like to watch. Well, this is because of AI which you tube uses in different ways.
Some of the ways are –
- Automatically remove the irrelevant content: Youtube algorithms remove those videos that are not relevant or have objectionable content. Around 8.3 million videos were removed that have irrelevant content in the first four months of this year. Around 70% were automatically identified by the AI and take down the videos before any views by the user.
This is the real power of machine learning once they learn the algorithm, they can do any task within a second on a huge amount of data which human even can’t imagine to do that fast as AI does.
The main aim of youtube adopt AI is to stop that type of content on it that leads to giving wrong information or content violation messages and spread something that is wrong.
Although AI do the things very properly but sometimes it pulled down the newsworthy videos mistakenly seeing it as “violent extremism”. That’s why Google have full-time human specialist that work with AI to address violative content.
- Recommended videos when you have done with a video, or you open the homepage of youtube:
Youtube recommends systems are the most common forms of the Machine Learning that users will encounter daily. Slowly youtube move towards the Deep learning for the better recommendation systems. According to some information available in the market, youtube is the second most visited website in the US, with an average of 400 hours of content uploaded per minute. In their paper, Covington et al. demonstrate a two-stage information retrieval approach, where one network generates the recommendation, and the other one ranks these generated recommendations. This approach is quite good as having only one network for doing this two alone was not giving an accurate answer.
- Automatically generates playlists of songs
- Advertising alongside videos
There were two main factors behind You Tube’s Deep Learning approach towards Recommender Systems:
Source: Deep learning pdf by YOUYUBE
There are two types of network uses by the You Tube are –
- The Generation Network
- Ranking the Predictions Network
The Generation Network
It is used to take the user’s activity history like IDs of the videos being watched, user history, liked videos, etc., and output the hundreds of the videos that might broadly applicable to the user.
This network is used to optimize the youtube according to the user interest.
Covington et al. poses the candidate generation problem as an extreme multiclass classification problem, where the prediction problem becomes “accurately classifying some watch time at w(t) at some time t”, for some given item i, context C, and user U.
where u belong to the R to the power N represents a high-dimensional “embedding” of the user, context pair and the v(j) belongs to the R to the power N represent embedding of each candidate video.
Ranking the Predictions Network
The candidate generation model has access to features such as video embedded, and the number of watches. The ranking network can take features such as thumbnail images and the interest of their peers in order to provide more precise and accurate scoring.
The objective of making this network is to maximize the expected watch time of the given recommendation of the video.
To predict the expected watch time the authors used logistic regression. It has probability E[T](1+P), where E[T] models the expected watch time of the impression, and P models the probability of clicking the video.
Youtube exhibits the impact of a wider and deeper network on per-user loss. Per-user loss is the total mispredicted watch time against the total watch time on held-out data. This allows the model to predict the data of per-user loss and show the good recommendation instead of predicting by itself.
“Machine learning changes video creation by the creators on YouTube”
In earlier when youtube was new at that time it count the higher clicks and recommend user for watching it. Further, Youtube introduces the whole machine learning and deep learning to personalize the user experience.
In 2020 creators who want to grow on YouTube have a deep knowledge of how and in what way the Youtube algorithm works. The creators must have to do the below-mentioned things as to grow on YouTube –
- Attract viewers with the right Title and Thumbnail:
- Choose video titles that inform what your video is about
- Provide translate options for titles, descriptions, and captions
- Descriptive thumbnail that summarizes your video content in short.
- Organize and program your content
- Create long watch-time videos.
- Create regular videos if you are new to this platform.