Multi-Framework Support in TensorFlow Federated

TensorFlow Federated (TFF) has been designed to support a broad range of federated computations, expressed through a combination of TFF's federated operators that model distributed communication, and local processing logic.

Currently local processing logic can be expressed using TensorFlow APIs (via @tff.tf_computation) at the frontend, and is executed via the TensorFlow runtime at the backend. However, we aim to support multiple other (non-TensorFlow) frontend and backend frameworks for local computations, including non-ML frameworks (e.g., for logic expressed in SQL or general-purpose programming languages).

In this section, we'll include information on:

• Mechanisms that TFF provides to support alternative frameworks, and how you can add support for your preferred type of frontend or backend to TFF.

• Experimental implementations of support for non-TensorFlow frameworks, with examples.

• Tentative future roadmap for graduating these capabilities beyond the experimental status.