Coefficient Complexity in Low-Access Quantized Linear Computations

Vinayak Ramkumar*, Netanel Raviv, Itzhak Tamo

*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

Quantizing linear computations over the reals is a common practice in many distributed systems, with applications in efficient, private, or robust machine learning algorithms. Recent work studied the access parameter of such computations, i.e., the maximum number of data entries required to accommodate any given computation. It was shown that lower access can be achieved with higher redundancy and vice versa, and a corresponding scheme based on covering codes was given. Surprisingly, the resulting scheme is universal to all possible two-valued quantizations, meaning that the same storage protocol can be used to retrieve any linear combination with two distinct coefficients - regardless of what those coefficients are - with the same access parameter. In this paper we extend this universal protocol to all possible quantizations with any number of values; while the storage parameter remains identical, the access parameter increases according to a new additive-combinatorics property we call coefficient complexity. We then turn to study the coefficient complexity - we characterize the complexity of small sets of coefficients, provide bounds, and identify coefficient sets having the highest and lowest complexity. Interestingly, arithmetic progressions have the lowest possible complexity, and some geometric progressions have the highest possible complexity, the former being particularly attractive for its common use in uniform quantization.

Original languageEnglish
Title of host publication2023 59th Annual Allerton Conference on Communication, Control, and Computing, Allerton 2023
PublisherInstitute of Electrical and Electronics Engineers Inc.
ISBN (Electronic)9798350328141
DOIs
StatePublished - 2023
Event59th Annual Allerton Conference on Communication, Control, and Computing, Allerton 2023 - Monticello, United States
Duration: 26 Sep 202329 Sep 2023

Publication series

Name2023 59th Annual Allerton Conference on Communication, Control, and Computing, Allerton 2023

Conference

Conference59th Annual Allerton Conference on Communication, Control, and Computing, Allerton 2023
Country/TerritoryUnited States
CityMonticello
Period26/09/2329/09/23

Keywords

  • Access-Redundancy
  • coded computation
  • covering codes
  • distributed systems

Fingerprint

Dive into the research topics of 'Coefficient Complexity in Low-Access Quantized Linear Computations'. Together they form a unique fingerprint.

Cite this