Abstract
Due to the high communication cost in distributed and federated learning, methods relying on compressed communication are becoming increasingly popular. Besides, the best theoretically and practically performing gradient-type methods invariably rely on some form of acceleration/momentum to reduce the number of communications (faster convergence), e.g., Nesterov’s accelerated gradient descent [31, 32] and Adam [14]. In order to combine the benefits of communication compression and convergence acceleration, we propose a compressed and accelerated gradient method based on ANITA [20] for distributed optimization, which we call CANITA. Our CANITA achieves the first accelerated rate O (r(1 + qωn3)Lǫ + ω(1ǫ)31) , which improves upon the state-of-the-art non-accelerated rate O ((1 + ωn)Lǫ + ωω2++nω 1ǫ ) of DIANA [12] for distributed general convex problems, where ǫ is the target error, L is the smooth parameter of the objective, n is the number of machines/devices, and ω is the compression parameter (larger ω means more compression can be applied, and no compression implies ω = 0). Our results show that as long as the number of devices n is large (often true in distributed/federated learning), or the compression ω is not very high, CANITA achieves the faster convergence rate O(qLǫ ) , i.e., the number of communication rounds is O (qLǫ ) (vs. O(Lǫ ) achieved by previous works). As a result, CANITA enjoys the advantages of both compression (compressed communication in each round) and acceleration (much fewer communication rounds).
Original language | English (US) |
---|---|
Title of host publication | Advances in Neural Information Processing Systems 34 - 35th Conference on Neural Information Processing Systems, NeurIPS 2021 |
Editors | Marc'Aurelio Ranzato, Alina Beygelzimer, Yann Dauphin, Percy S. Liang, Jenn Wortman Vaughan |
Publisher | Neural information processing systems foundation |
Pages | 13770-13781 |
Number of pages | 12 |
ISBN (Electronic) | 9781713845393 |
State | Published - 2021 |
Event | 35th Conference on Neural Information Processing Systems, NeurIPS 2021 - Virtual, Online Duration: Dec 6 2021 → Dec 14 2021 |
Publication series
Name | Advances in Neural Information Processing Systems |
---|---|
Volume | 17 |
ISSN (Print) | 1049-5258 |
Conference
Conference | 35th Conference on Neural Information Processing Systems, NeurIPS 2021 |
---|---|
City | Virtual, Online |
Period | 12/6/21 → 12/14/21 |
Bibliographical note
Publisher Copyright:© 2021 Neural information processing systems foundation. All rights reserved.
ASJC Scopus subject areas
- Computer Networks and Communications
- Information Systems
- Signal Processing