iFLearner is a powerful and lightweight federated learning framework that provides a computing framework based on data privacy security protection, mainly for federated modeling in deep learning scenarios. Its security bottom layer supports multiple encryption technologies such as homomorphic encryption, secret sharing, and differential privacy. The algorithm layer supports various deep learning network models, and also supports mainstream frameworks such as Tensorflow, Mxnet, and Pytorch.

architecture

iFLearner is mainly designed based on the following principles:

• event-driven mechanism: Use the event-driven programming paradigm to build federated learning, which regards federated learning as a process of sending and receiving messages between participants, and describes the federated learning process by defining message types and the behavior of processing messages.

• Training framework abstraction: Abstract deep learning backend, compatible with Tensorflow, Pytorch and other framework backends.

• High scalability: Modular design, users can customize the aggregation strategy, encryption module, and support algorithms in various scenarios.

• lightweight and simple: The framework is at the Lib level and is lightweight enough. At the same time, users can simply transform their own deep learning algorithms into federated learning algorithms.