Compiler-Driven Approximation Tuning for Hyperdimensional Computing Conference: Preprint; ; 2026

Xavier Routh, Abdul Rafae Noor, Akash Kothari, Zheyu Li, Mahbod Afarin, Tajana Rosing, Vikram Adve · 2026

Read Compiler-Driven Approximation Tuning for Hyperdimensional Computing Conference: Preprint; ; 2026 by Xavier Routh, Abdul Rafae Noor, Akash Kothari, Zheyu Li, Mahbod Afarin, Tajana Rosing, Vikram Adve (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2606-26547.txt.

Description

Xavier Routh and six colleagues present ApproxHDC, a compiler framework for hyperdimensional computing. That approach represents data as very long vectors and computes with simple operations on them. It tolerates noise well and maps naturally onto varied hardware. Small losses in answer quality can buy large speed gains. Finding the best places to approximate, though, is a huge search problem.

ApproxHDC extends an existing compiler so one program can target many back ends. Targets include ordinary processors, graphics processors, and simulated memory-based accelerators. The space of software and hardware approximations grows exponentially, so exhaustive search is impossible. The framework uses efficient search and analysis to find high-impact approximations instead. The authors report substantial performance gains with minimal quality loss. The work aims to make approximation tuning automatic rather than a hand craft.

Catalog record

CREATORXavier Routh, Abdul Rafae Noor, Akash Kothari, Zheyu Li, Mahbod Afarin, Tajana Rosing, Vikram Adve
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE59 KB
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2606.26547

Browse the full library · About the Tower of Babel