“TensorFlow Federated (TFF) is an open-source framework for machine learning and other computations on decentralized data. TFF has been developed to facilitate open research and experimentation with Federated Learning (FL), an approach to machine learning where a shared global model is trained across many participating clients that keep their training data locally. For example, FL has been used to train prediction models for mobile keyboards without uploading sensitive typing data to servers.“
This has super broad applicability, beyond mobile, federated learning could become the answer to invasive, data hungry centralized applications that "need the data to make the system better"... no, you just need to learn locally & send a learning summary to a central system, not the data.
This recent paper from Google combines FL with secure multi-party computation for “secure aggregation” to prevent the aggregate server from even seeing the gradients https://export.arxiv.org/abs/1902.01046
3 comments
[ 0.18 ms ] story [ 18.3 ms ] threadFrom https://www.tensorflow.org/federated
This is a very exiciting project that compliments other privacy-preserving machine learning techniques (PPML) in TensorFlow:
- Differential Privacy: https://github.com/tensorflow/privacy
- Secure Multi-Party Computation: https://github.com/mortendahl/tf-encrypted
- Confidential Computing (Trusted Execution Environments, Asylo, Intel SGX): https://github.com/dropoutlabs/tf-trusted
This recent paper from Google combines FL with secure multi-party computation for “secure aggregation” to prevent the aggregate server from even seeing the gradients https://export.arxiv.org/abs/1902.01046