| Title: |
Catalyst Acceleration of Error Compensated Methods Leads to Better Communication Complexity |
| Authors: |
Qian, Xun; Dong, Hanze; Zhang, Tong; Richtarik, Peter |
| Contributors: |
King Abdullah University of Science and Technology Thuwal, Saudi Arabia; Computer Science Program; Visual Computing Center (VCC); Computer, Electrical and Mathematical Science and Engineering (CEMSE) Division; Shanghai Artificial Intelligence Lab Shanghai, China; The Hong Kong University of Science and Technology Hong Kong |
| Publisher Information: |
ML Research Press |
| Publication Year: |
2023 |
| Collection: |
King Abdullah University of Science and Technology: KAUST Repository |
| Description: |
Communication overhead is well known to be a key bottleneck in large scale distributed learning, and a particularly successful class of methods which help to overcome this bottleneck is based on the idea of communication compression. Some of the most practically effective gradient compressors, such as TopK, are biased, which causes convergence issues unless one employs a well designed error compensation/feedback mechanism. Error compensation is therefore a fundamental technique in the distributed learning literature. In a recent development, Qian et al (NeurIPS 2021) showed that the error-compensation mechanism can be combined with acceleration/momentum, which is another key and highly successful optimization technique. In particular, they developed the error-compensated loop-less Katyusha (ECLK) method, and proved an accelerated linear rate in the strongly convex case. However, the dependence of their rate on the compressor parameter does not match the best dependence obtainable in the non-accelerated error-compensated methods. Our work addresses this problem. We propose several new accelerated error-compensated methods using the catalyst acceleration technique, and obtain results that match the best dependence on the compressor parameter in non-accelerated error-compensated methods up to logarithmic terms. ; The work of Xun Qian and Peter Richtárik was supported by the Extreme Computing Research Center at KAUST, and the work of Peter Richtárik was also partially supported by the SDAIA-KAUST Center of Excellence in Data Science and Artificial Intelligence. |
| Document Type: |
conference object |
| File Description: |
application/pdf |
| Language: |
unknown |
| ISSN: |
2640-3498 |
| Relation: |
https://proceedings.mlr.press/v206/qian23a.html; 2301.09893; 2-s2.0-85165128216; 615-649; http://hdl.handle.net/10754/687397; 206 |
| Availability: |
http://hdl.handle.net/10754/687397 |
| Rights: |
Copyright © The authors and PMLR 2023. MLResearchPress. Archived with thanks to ML Research Press. |
| Accession Number: |
edsbas.BE53DA48 |
| Database: |
BASE |