As technology advances, AI is becoming more and more integrated into our lives. Numerous industries and companies are using AI to improve their products and services. For instance, Google uses machine learning to build personalized services and maximize user experiences. Unfortunately, it also comes with a challenge as such models need to be trained on tons of personal data in order to perform well. This certainly becomes an issue as we are increasingly concerned about privacy. Here’s where federated learning comes to play.
In this blog post, I will introduce the concept of federated learning, why it matters, and how it could potentially shape the future of AI.
What is federated learning?
Federated learning was first introduced by Google in 2017, in a blog post titled “Federated Learning: Collaborative Machine Learning without Centralized Training Data”. As they put it,
“Federated learning enables mobile phones to collaboratively learn a shared prediction model while keeping all the training data on device, decoupling the ability to do machine learning from the need to store the data in the cloud.”
Thus, federated learning can be thought of as a new approach to machine learning that allows you to train models across devices without pooling the data. Instead of bringing users’ data to the server, federated learning brings the machine learning model to the users’ devices.
With the above idea in mind, the way federated learning works is actually quite intuitive. A general federated learning framework usually involves the following steps:
- Your device downloads the current model from the server and improves it by training from data on your device
- Your device uploads the training results in the form of a small focused update to the server using encrypted communication
- The server averages the updates across all users to improve the shared model
Here, it is worth noting that federated learning is an iterative process; which means that the above steps will be repeated until a pre-defined termination criterion is met (e.g., a maximum number of iterations is reached or the model accuracy is greater than a threshold).
Note: If you’re interested to learn more about how federated learning works on a technical level, you can check out this 2016 paper published by Google AI researchers. If you prefer a less technical overview, you should check out this online comic instead.
Why does federated learning matter?
Traditional machine learning adopts a centralized approach which requires all the data to be brought together to a server, where the models are trained. This centralized training approach, however, is privacy-intrusive as users have to trade their privacy by sending their personal data to the server owned by the AI companies.
In light of this, federated learning is basically the decentralized form of machine learning. Unlike the traditional machine learning approaches, federated learning does not require the data to be stored in the server. Instead, the data stays on each users’ device and thus privacy is preserved. In such a case, users can benefit from obtaining a well-trained machine learning model without sending their sensitive personal data to the server.
How could federated learning shape the future of AI?
In recent years, data privacy has become the major obstacle for the development of AI, especially with the establishment of data protection and privacy laws like the General Data Protection Regulation (GDPR) and the California Consumer Privacy Act (CCPA). Thanks to federated learning, these barriers have been broken and we now have a secure way of training machine learning models within users’ devices.
Federated learning allows us to build better machine learning models that can provide personalized services and maximize user experiences. A huge amount of data are produced from user’s devices on a daily basis. These data are valuable as it contains personal information about the users and their personal interests. With federated learning, it is now possible to build such personalized models while still preserving users’ privacy.
What's Next?
In this blog post, we have explored a new decentralized approach to machine learning called federated learning. I personally believe that federated learning will play an important part in shaping the future of AI. In the near future, we will see a plethora of new applications taking advantage of federated learning, enhancing user experience in a way that was not possible before. Of course, this breakthrough also comes with a unique set of challenges that AI researchers need to tackle to bring this field forward. Nevertheless, the future looks exciting if AI and privacy could go hand in hand.