I saw a series of learning about federated learning, which is quite interesting. I will record it here, and I will attach the original address at the end.
Article directory
1. Comics display
This is a cartoon about federated learning.
Two, about
1. Manga address
Comic address: https://federated.withgoogle.com
This website is made by the federated learning team of Google artificial intelligence.
Story Content Creation: Lucy Bellwood, Scott McCloud
Art Creation: Lucy Bellwood
This comic is licensed under the Creative Commons Attribution Non-Commercial NoDerivative Works 3.0 License.
Translations are allowed.
2. Translation of specific content of comics
One day, Xiaomei came back from the meeting and excitedly told her boss Sanmao: "Boss, I know how to win back our customers!"
The boss Sanmao said with a sad face: "It's overdue, now our brand is a mess."
Xiaomei said confidently: "I have learned a new way to deal with our privacy issues and improve functions! That is federated learning."
Sanmao is confused, what is federated learning?
"It allows users to do machine learning while saving data on the device. It's resilient, low-impact, and safe—" Xiaomei was interrupted by her boss before she could finish her sentence.
"Very good, I am convinced by you, I will give you our best team!", the boss excitedly went to find someone, "Of course it is an intern~" Xiaomei was speechless for a moment, "
Okay Come on, accept the challenge."
"Hey, everyone, let's start with a hypothetical problem: we need to train machine learning models on user privacy data ." Xiaomei said straight to the point.
(A series of fish-catching actions in the middle)
Xiaomei pointed out: "The real performance of a machine learning model depends on the relevance of the data used to train it. And the best data is stored in various devices that are used every day. "
But the intern Xiaolan asked: "Isn't accessing those sensitive data the whole reason your application crashes?" "
Yes, that's right, but what if the data is not passed out?" Xiaomei said.
"Then I can't train the model!" said the intern vigorously.
"No, you can." Xiaomei said mysteriously, "Welcome to the world of federated learning!" "
We will train a centralized model on decentralized data, and the data on the device can train a smarter center model, which makes our user experience better." "
But because the data does not come out of the device, we transfer the model to the personal device, and not all devices are mandatory to participate, only when the device meets the conditions to participate, so that users and Devices are not affected by model training."
After local training, these device subsets will send some parameter results after training, and in order to prevent the server from reconstructing data or models through these uploaded parameters, these uploaded data are encrypted. , and can be encrypted with a key that the server does not have.
...
Finally, when everyone reached an agreement, the boss came in and said that the company was completely bankrupt!
Then everyone happily formed a new company, except the boss, lol
Summarize
This cartoon is to point out the advantages of federated learning from the point of view of people's worries. It is worth reading.