How do you leverage AI and at the same time keep control of your data?
A report from the Data Innovation Summit 2019, and our session about federated machine learning.
Data Innovation Summit is a yearly gathering, fourth edition this year, with over 2000 visitors and 100+ Nordic and international data innovation companies presenting on six stages covering topics such as:
- Data science and engineering
- Machine Learning
- Artificial Intelligence
- Applied Innovation
- Analytics and Visualisation
- Data Management
Scaleout had two sessions. On day one with Daniel Zakrisson on the machine & deep learning stage, and day two with Ola Spjuth on the data engineering stage.
The current paradigm in machine learning
Normally, machine learning starts with collecting as much data as possible in a central datastore, then develop machine learning models on it.
Often data cannot be moved away from its source. There are three main reasons for this:
- Private/Proprietary Data — Sharing valuable business data with someone else is not an option.
- Regulated Data — GDPR, medical data, etc.
- Practical blockers — data is too big, the network connection is expensive, slow or unreliable.
The result is that only a fraction of the available data is currently used and therefore inhibiting machine learning models reaching its full potential.
How can organisations collaborate on building machine learning models without giving up ownership of their data?
A lot could be gained if there was a way for organisations to collaborate on building machine learning models. But how can it be done without risking to transfer ownership of your data, or leaking regulated, private information?
The area of machine learning focused on this problem is federated machine learning. It is simple in concept: data never moves, instead local machine learning models are trained on the data.
Federated learning makes it possible to build machine learning systems without moving the training data. The data remains in its original location, which helps to ensure privacy and reduces communication complexity and costs.
To further increase privacy and security, other methods can be used in combination with federated learning. Differential privacy and homomorphic encryption are methods that increase privacy and can be applied in both regular machine learning case, and in federated machine learning.
Federated machine learning is still bleeding edge research and is not yet operationalised and off the shelf. Scaleout and a few other front runners are working on it in different areas and we are committed to bridging the gap between federated machine learning research and production-grade systems.
Details about federated machine learning in our presentation below. Get in touch if you want to learn more!
Scaleout has an experienced team from both the industry and leading academic researchers in applied machine learning, cloud and fog computing, and scientific computing from Uppsala University.
Follow us on Linkedin for updates of our latest research on federated machine learning:
Visit the Scaleout site for more information about the Lean AI — a structured and easy-to-follow process for taking AI to production with the goal of delivering business value quickly and efficiently.
