WHAT'S HIS NAME ?

An AI project by Avi JAOUI

You have this actor's name on the tip of your tongue ?
Here, let me help you out with that !

Before I can help you finding the name of the actor you desperately want to know because you told your friends : "I swear it's this guy ! He plays in... sh*t I can't find it but I swear it's him !", let's talk about Artificial Intelligence, or AI. Artificial Intelligence is a something many people, many companies and many governments have been working on for years, and it is thought to be of great help for everyone.
One of the many tasks AI is linked to is machine learning, which is the concept of a program learning autonomously thanks to the databases it has been given access to. For example, it is possible to have an AI recognize images, by a process called deep learning.
If you provide a program with a wide range of pictures of the same item, let's say a teapot, the program will be able, after processing the data, to recognize a teapot when it is shown another picture of one.
teapot


So how does it work ? Each time we give the AI an image, the program learns and stocks the image. When the AI is given a new picture, the program will run an alorithm which will allow it to analyse the pictures that are already in stock and the new picture. The more we show the AI pictures, the more its ability to recognize items is accurate.

Many tools can be used to learn how to implement AI into a web page, like MobileNetsand Ml5.js. The first one are programs which are trained to identify and recognize items in pictures, and the second one is a database providing access to algorithms.

code
We have here an example of code which is written in order to use Ml5.js and MobileNets.



What is this actor's name ?


Imagine you are with your friends at home for a movie night, and you chose to watch an old movie, Firstborn, from 1984. You recognize actors like Terri Garr or Peter Weller, but you can't put your fingers on the name of the actor playing that teenager. Well, this is a job for Artificial Intelligence.

First, so as to obtain an AI program capable of identifying the name of an actor from a screenshot of a movie, whatever the year, you need to feed the program a great amount of pictures of this actor.


Here are a few pictures :


All these pictures are used to create an image database. Then, when you use MobileNets, the program will try to identify the new picture by giving it a degree of resemblance, calculated with a percentage of confidence which is based on the classification made into the database.


Let's input the screenshot of the actor into the program.

MobileNet is 73.42% sure this is Robert Downey Jr.

Here, the Ai is confident at 73.42%, which means the program needs more pictures of Robert Downey Jr, from all of his movies, so as to create database powerful enough to be 100% accurate when recognizing the actor.

Now, we will try with a picture of Robert Downey Jr. starring in Chaplin, where he plays Charlie Chaplin and has to look like him.



MobileNet is 39.86% sure this Robert Downey Jr.

In this case, the AI truly unsure about the answer. This is because in this picture, Robert Downey Jr. looks a lot like the real Charlie Chaplin and not so much like the other pictures of him. If the AI also has pictures of Charlie Chaplin in its databse, as its goal is to identify any actor, then there could have been a mix-up between both actors.

Think of this AI next time you watch a movie, and try to learn as your machine does !