OpenMined in one line: OpenMined gives data scientists a way to use sensitive data to train machine learning models while protecting data privacy.
OpenMined is an open-source community that protects privacy in machine learning via two components: the PySyft Python library and PyGrid hosting platform. These tools give data scientists a way to use sensitive data for training and analysis without having direct access to it – the data stays on its owners' local machines.
The PySyft library is an extension for PyTorch, TensorFlow, and Keras which allows data scientists to train models on remote devices using encrypted computation.
PyGrid is a peer-to-peer model hosting platform. Distributed devices, each hosting local data, can download a model hosted on PyGrid for local training before aggregating these distributed models to produce a global, trained model.
The data required to train machine learning models is sometimes sensitive and should remain private, but the standard for training models requires data to be centralized. This norm poses a problem for both data owners and data scientists; the former may not be able to give up control of data if it contains sensitive information about individuals, while the latter cannot train models without access to data.
OpenMined solves this problem by separating model creation and training. Data can remain distributed on the owners' devices to keep sensitive information private, and data scientists can train models without directly accessing the data. This system provides privacy to the data owner and also assists in scaling the training process through distributed computation.