On the Convergence of Local SGD on Identical and Heterogeneous Data (Pilot talk of FLOW: Federated Learning One World Seminar)

The Federated Learning One World (FLOW) seminar provides a global online forum for the dissemination of latest scientific research results in all aspects of federated learning, including distributed optimization, learning algorithms, privacy, cryptography, personalization, communication compression, and new generation models.

The talks will address the theoretical foundations of the field, applications, datasets, benchmarking, software and systems.

Hosts:

• Peter Richtárik, KAUST, Saudi Arabia (Chair)
• Virginia Smith, Carnegie Mellon, USA
• Aurélien Bellet, Inria, France
• Dan Alistarh, IST, Austria

Access the talk slides.

Recording of the talk:

https://www.youtube.com/watch?v=6ThWeKQyp8k&feature=emb_logo