Scaleout Federated Platform
Our current priority at Scaleout is to build our Federated platform to bridge the gap between academic research on privacy-preserving learning on the one hand, and the requirements of production grade systems for large scale applications on the other hand. Our vision is a machine learning framework-agnostic platform that removes the technical, security and trust-related barriers to forming an ML-alliance, and that lets data scientists focus on building federated models.
The current paradigm
In the current paradigm of centralized machine learning, developing accurate models typically starts with collecting as much data as possible in a central data store, then develop machine learning models on the collected, pooled data.

Centralized machine learning is by far the most common practice but it comes with several challenges that diminish its potential in AI systems. Only a fraction of the possible available data is currently accessible and therefore inhibiting machine learning models reaching its full potential.
Generally speaking, more data means better machine learning models, but in many instances, data cannot be gathered in a central location.

Reasons for this include:
  • Private/Proprietary data — Sharing valuable business data with someone else is not an option.
  • Regulated data — GDPR, HIPAA, etc.
  • Practical blockers — data is too big, the network connection is expensive, slow or unreliable.
Main issues with centralized machine learning
Federated Machine Learning
Federated Machine Learning (FedML) is a distributed machine learning approach which enables training on decentralised data. A server coordinates a network of nodes, each of which has local, private training data. The nodes contribute to the construction of a global model by training on local data , and the server combines non-sensitive node model contributions into the global model.
Benefits of FedML
Federated learning addresses the fundamental problems of centralized AI such as privacy, ownership, and locality of data. It extends, even disrupts, the centralized AI paradigm in which better algorithms always comes at the cost of collecting more and more sensitive data.

Federated learning enables:
  • Data security and privacy where data never moves
  • Reduced communication complexity and costs
  • Powerful data network effects in industries where data cannot be transferred
How it works?
In simple terms, it is an approach where you "bring the code to the data, instead of the data to the code".
For example, in federated averaging:
  • Nodes receive the latest model from the server and start training locally.
  • Every set number of steps the nodes sends partially trained models to the server.
  • The server combines the models from the local nodes into a global federated model
  • The federated model is sent to the nodes and cycle starts over.
Factors influencing the FedML strategy
Other ways to create global models include strategies related to ensemble models. In the end, the best strategy to construct a global model depends on a number of factors including:

  • What can be shared? (training data, test/validation data, model parameters or predictions only)
  • The problem and type of model (CNN, SVM, Random Forest etc)
  • The local training cost and network communication cost
  • The target size of the data alliance
  • The need for valid prediction intervals or set predictions
  • Is the target prediction endpoints in embedded systems or in SaaS?
For this reason we aim for our solutions to be flexible and ML framework agnostic supporting sustainable development of AI.
Interested? Let us know!
We believe that federated machine learning holds exceptional promise. The reason is that it has equal potential for constructing integrity-preserving machine learning models at scale, and for enabling powerful business models centred around data-alliances.
Feel free to reach out to us so we can discuss more!
Phone:
+46 (0) 70-880 15 70

Mail:
contact@scaleoutsystems.com


By clicking the button you agree with our Privacy Policy.