On the heels of TensorFlow Privacy, a library for training AI models with “strong privacy guarantees,” Google today debuted another module for its TensorFlow machine learning framework: TensorFlow Federated (TFF). It’s intended to make it easier to experiment with machine learning and other computations on decentralized data, according to the Mountain View company, by implementing an approach called federated learning.
It follows hot on the heels of TensorFlow 2.0 alpha, which was also announced today.
“We have designed TFF based on our experiences with developing the federated learning technology at Google, where it powers ML models for mobile keyboard predictions and on-device search,” Google wrote in a Medium post published this morning. “With TFF, we are excited to put a flexible, open framework for locally simulating decentralized computations into the hands of all TensorFlow users.”
Here’s the crux of it: TFF enables developers to deploy an AI system and train it across data from multiple sources, all while keeping each of those sources separate and local. It comes with an API — Federated Core (FC) API — that supports a “broad range” of computations over a decentralized dataset, and which expresses a new data type that specifies both the underlying data and the location of that data on distributed clients.
TFF also includes a local machine runtime that simulates the computation being executed across a set of data-holding clients, with each client computing their local contribution and with a centralized coordinator aggregating all contributions. But from a developer perspective, Google says, the federated computation is basically a function with inputs and outputs that happen to reside in different places.
“With TensorFlow Federated, we are taking a step towards making the technology accessible to a wider audience, and inviting community participation in developing federated learning research on top of an open, flexible platform,” Google said. “Over time, we’d like TFF runtimes to become available for the major device platforms.”