Members-Only
Recent Talks & Demos are for members only
You must be an AI Tinkerers active member to view these talks and demos.
Flame: Federated Learning Framework
This talk covers federated learning using Project Flame, demonstrating decentralized model training on edge devices to enhance privacy and reduce computation.
Traditional machine learning depends on the centralization of data, but that comes with privacy and computational concerns. A reality with billions of edge devices diminish those issues, especially with the advent of federated machine learning. Training may be performed on edge devices directly, keeping datasets decentralized and private. Additionally, offloading work to different nodes means less computation per device. Projects like GBoard, Siri, and even the medical and military fields already use federated learning.
One current open-source framework for federated machine learning is Project Flame (maintained by Cisco Systems). Flame uses object-oriented programming to implement different graphs between edge devices for a federated learning network. Flame can be extended to different kinds of topologies and executed across multiple devices that run Python code using P2P communication.
Flame provides scalable federated learning for edge, leveraging LIFL architecture.