Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Improved Data Encoding for Emerging Computing Paradigms: From Stochastic to Hyperdimensional Computing

Mehran Shoushtari Moghadam, Sercan Aygun, M. Hassan Najafi

TL;DR

This paper addresses data encoding in SC and HDC, where randomness quality critically impacts accuracy and hardware cost. It introduces a hardware-friendly deterministic RNG based on low-discrepancy Van der Corput sequences with bases $2^n$ (VDC-$2^n$) to generate bit-streams and hypervectors with improved correlation properties. The approach yields substantial gains in SC and HDC performance, including improved accuracy and energy efficiency, and enables efficient edge deployment through hardware-lean implementations and UnaryHD architectures. Empirical results across SC non-linear functions and HDC image/classification benchmarks demonstrate notable improvements over baselines and highlight the potential for scalable, resource-constrained AI systems.

Abstract

Data encoding is a fundamental step in emerging computing paradigms, particularly in stochastic computing (SC) and hyperdimensional computing (HDC), where it plays a crucial role in determining the overall system performance and hardware cost efficiency. This study presents an advanced encoding strategy that leverages a hardware-friendly class of low-discrepancy (LD) sequences, specifically powers-of-2 bases of Van der Corput (VDC) sequences (VDC-2^n), as sources for random number generation. Our approach significantly enhances the accuracy and efficiency of SC and HDC systems by addressing challenges associated with randomness. By employing LD sequences, we improve correlation properties and reduce hardware complexity. Experimental results demonstrate significant improvements in accuracy and energy savings for SC and HDC systems. Our solution provides a robust framework for integrating SC and HDC in resource-constrained environments, paving the way for efficient and scalable AI implementations.

Improved Data Encoding for Emerging Computing Paradigms: From Stochastic to Hyperdimensional Computing

TL;DR

This paper addresses data encoding in SC and HDC, where randomness quality critically impacts accuracy and hardware cost. It introduces a hardware-friendly deterministic RNG based on low-discrepancy Van der Corput sequences with bases (VDC-) to generate bit-streams and hypervectors with improved correlation properties. The approach yields substantial gains in SC and HDC performance, including improved accuracy and energy efficiency, and enables efficient edge deployment through hardware-lean implementations and UnaryHD architectures. Empirical results across SC non-linear functions and HDC image/classification benchmarks demonstrate notable improvements over baselines and highlight the potential for scalable, resource-constrained AI systems.

Abstract

Data encoding is a fundamental step in emerging computing paradigms, particularly in stochastic computing (SC) and hyperdimensional computing (HDC), where it plays a crucial role in determining the overall system performance and hardware cost efficiency. This study presents an advanced encoding strategy that leverages a hardware-friendly class of low-discrepancy (LD) sequences, specifically powers-of-2 bases of Van der Corput (VDC) sequences (VDC-2^n), as sources for random number generation. Our approach significantly enhances the accuracy and efficiency of SC and HDC systems by addressing challenges associated with randomness. By employing LD sequences, we improve correlation properties and reduce hardware complexity. Experimental results demonstrate significant improvements in accuracy and energy savings for SC and HDC systems. Our solution provides a robust framework for integrating SC and HDC in resource-constrained environments, paving the way for efficient and scalable AI implementations.
Paper Structure (5 sections, 3 figures, 4 tables)

This paper contains 5 sections, 3 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Background and Related Work
  3. Proposed Framework
  4. Results and Contributions
  5. Conclusion

Figures (3)

  • Figure 1: Pseudo-randomness vs. Quasi-randomness in random sequences. (a),(b) Points Distribution plots, (c),(d) Value Spread plots, (e),(f) Correlation/Orthogonality plots, and (g), (h) Design Implementation.
  • Figure 2: Applying our Deterministic approach to the encoding part of SC and HDC models. (a) Implementing non-linear functions and division operation using the proposed VDC-2$n$ sequence generator in SC and (b) Improving HDC performance by generating high-quality $\mathcal{HV}$s using deterministic sequences.
  • Figure 3: Performance evaluation of the end-to-end Unary structure for HDC on the DermaMNIST dataset Yang2023 using single-source $\mathcal{HV}$ generator design (Fig. \ref{['designs']}(b)➑), D=1024 Mehran-islped.