Detailed comparison of communication efficiency of split learning and federated learning

Подробное сравнение эффективности коммуникации при раздельном и федеративном обучении
Ramesh Raskar, Otkrist Gupta, Praneeth Vepakomma, Abhishek Singh
2019-09-18

communication efficiencydistributed machine learningfederated learningmodel size and client countsplit learning
We compare communication efficiencies of two compelling distributed machine learning approaches of split learning and federated learning. We show useful settings under which each method outperforms the other in terms of communication efficiency. We consider various practical scenarios of distributed learning setup and juxtapose the two methods under various real-life scenarios. We consider settings of small and large number of clients as well as small models (1M - 6M parameters), large models (10M - 200M parameters) and very large models (1 Billion-100 Billion parameters). We show that increasing number of clients or increasing model size favors split learning setup over the federated while increasing the number of data samples while keeping the number of clients or model size low makes federated learning more communication efficient.
1
Federated learning is more communication-efficient when the number of data samples increases while client count and model size remain low.
2
Increasing model size favors split learning, across models ranging from 1M–6M to 1B–100B parameters.
3
Neither method universally dominates; communication efficiency depends on client population, model size, and data-sample volume.
4
Split learning becomes more communication-efficient than federated learning as the number of clients increases.
5
The paper systematically compares the communication efficiency of split learning and federated learning across practical distributed-learning scenarios.

Split learning and federated learning in distributed machine-learning setups

Communication efficiency under varying numbers of clients, model sizes, and data-sample counts

Publication Details
Publication Date
2019-09-18
Journal
Publisher
ISSN
Cited by
217
Access Type
Author Information
Authors
Ramesh Raskar
Otkrist Gupta
Praneeth Vepakomma
Abhishek Singh
Explore further
Open the scid.ai AI chat with a ready-made request: it will find papers on a similar topic and help build a literature review.
Find similar papers in the chat →
Make a presentation
100%