Last December, the Microsoft AI, ML Community hosted a ‘live’ online event where the Federated Learning team in AI Singapore walked attendees through the Synergos platform they had been building for federated learning. The high level architecture of the platform had already been outlined in a previous post. This event provided an opportunity for the team to explain to the community in greater detail what federated learning is in general and how Synergos works in particular. It was also a great chance to receive feedback and answer questions.
The event has been recorded and below is a list of the highlights. You can jump to the part which interests you by clicking on it.
- The Data Privacy Problem in Machine Learning
- How Federated Learning Solves the Data Privacy Problem
- Building Synergos
- Use Cases
- The Federation Component
- The Roadmap
You can also view the full recording below. Many thanks to host Setu Chokshi for making this possible.
Get in conversation with us at email@example.com. Let’s do great things together! 🤝
1. The Data Privacy Problem in Machine Learning
Machine learning requires immense amounts of data. However, this might not be directly accessible in certain applications for various reasons, including personal data protection concerns and business considerations.
2. How Federated Learning Solves the Data Privacy Problem
Federated learning is based on the principle of sharing how to adjust and improve a model, rather than the data upon which the model is trained.
3. Building Synergos
4. Use Cases
Go through some possible use cases for federated learning. Understand the steps involved and see the benefit of federated learning compared with the results from local training and centralized training in a pilot of Synergos in the healthcare domain to predict ICU in-hospital mortality.
5. The Federation Component
The Federation component in Synergos is where the federated learning takes place, which happens in three phases consisting of nine steps.
Using the heart disease public dataset from the UCI machine learning repository, the setup, model training and inference with 3 workers coordinated by a TTP (Trusted Third Party) is demonstrated. This dataset is used for demo here because: (1) it is a good use case for federated learning as the data is collected from multiple geographically distributed hospitals; and (2) the data is relatively small so that we can see a full cycle of federated learning within a short demo.
7. The Roadmap
See what features for Synergos are planned as part of the journey towards its official launch and beyond.
The Federated Learning Series
- AI Singapore’s Journey Into the World of Federated Learning
- Voices from the AISG Federated Learning Lab
- A Peek into Synergos – AI Singapore’s Federated Learning System
- Do Great Things Together With Federated Learning (this article)