Sharing ML-models in a setting where the raw data cannot be disclosed?!
Scaleout Systems will present their prototype for sharing machine-learning models using data alliances to preserve privacy and enable federation of learning from many assets.
Think of a fleet of assets used by different customers all learning via machine learning. Sharing the trained model is problematic since that might reveal a customers specific use of the asset. Still sharing the model and receiving models from others is crucial to improve the service of the asset.
Example, imagine a x-ray scanner with image recognition that improves its results by learning from other x-rays where you cannot share the patients medical record or condition.
Until then, read Daniel's post on medium https://medium.com/scaleout/sc...