Telephones, with the passage of time, they have been getting more and more smart . First came the Internet connection, then the applications, then the raw power and, finally, we are witnessing the emergence of the. Even if you do not notice it, your mobile learns from you with the passage of time . Learn how you write, how you behave, what usage patterns you have and that you use to adapt your mobile to your needs. But what if I told you that your mobile uses your data, and the data of millions of users, to become more and more intelligent? This, my friend, is called Federated Learning
First, before you start, you must understand How Machine Learning works . We will not go into details, but you could say that Machine Learning consists of a program / app / computer that learn by processing data provided in the form of examples. If I want to train a program to be able to perform facial recognition, I must train it with millions and millions of photos to learn what a face is, what it is like and what patterns it follows. The problem with this system is that requires that the data be centralized , that is to say, that there is an organism or entity that controls them.
This is what Google does with its cloud computing platform, one of the largest and most secure.
Federated Learning, or federated learning, allows a program, or in this case, a smartphone, learn from other terminals collaboratively , while keeping all the data in the device. Put another way, thanks to this technique, your phone will learn from the use that other people make of your mobile phone, and it will adapt to make your experience better. To see it with another perspective, We will approach this topic using GPS as an example.
Imagine that on this planet there were two types of people: some who use GPS a lot and others who never use it . Google would analyze the behavior patterns of both groups and, subsequently, modify the configuration of the devices to better adapt them to their usage trends . That way, Google would know that all the members of the first group use GPS a lot, so it makes sense for the GPS to update faster even if it consumes more battery power.
On the other hand, given that the members of the second group do not use GPS at all, it does not make sense that it is updated quickly, so the GPS refresh rate could be reduced in favor of achieving greater autonomy. In this way, the smartphones of the members of both groups they adapt and behave ideally for each user . Imagine this, but raised to the nth degree. Something like that is what Google does, only that with many algorithms and technology in between.
Do you find it interesting? As this is already running on the Google keyboard . Why Google knows what, when you write “pizzeria”, do you mean the Telepizza of your city? Because many people in your city have searched for “pizzeria” and have selected that Telepizza . Google has understood that when people write “pizzeria” they do it to find the address of that Telepizza, so, directly, He offers you his address because, possibly, it is what you are looking for . Learn from you and it makes your task easier.
This was already done , with the difference that the process was centralized. Smartphones had to send data to Google servers , where their algorithms analyzed and processed them. Not now. What is done in Federated Learning is Download a generic Machine Learning model that saves your records local way on the mobile . The way you use your terminal modifies this generic model, and what you do is send a “change log” . Basically, you’re telling Google how the model has changed, but you are not sending the raw data as such .
These records, that are anonymous , are sent – encrypted -, grouped and processed by Google to improve the Machine Learning model , which helps smartphones become smarter and smarter. This has two advantages:
So, thanks to Federated Learning and Machine Learning, we can achieve, among all, that Android is a smarter system and work for us . After all, what is technology but a tool to make life easier ? If you are interested in investigating in more complicated terms, Google itself published a post a while ago explaining how they had managed to develop this technology. You can read it.