XL-HD: Extended Learning in Hyperdimensional Computing via Deterministic Projections for In-Memory Accelerators

2026-05-23

AI Summary

This work addresses the challenge of deploying traditional hyperdimensional computing on resource-constrained edge devices, which typically relies on high-dimensional pseudorandom vectors and heuristic update rules. The authors propose the first end-to-end learnable, deterministic hyperdimensional computing framework that maps binary inputs to high-dimensional space using Sobol sequences, optimizes class prototypes in the real domain, and subsequently binarizes them to enable fully binary dot-product inference. The resulting architecture is compatible with ReRAM-based in-memory computing. Evaluated on MNIST, UCIHAR, and ISOLET benchmarks, the method achieves competitive accuracy while delivering exceptional hardware efficiency: the inference engine occupies only 0.395 mm² and consumes as little as 0.40 μJ per inference cycle, significantly improving both energy efficiency and hardware compactness.

Abstract

Hyperdimensional computing (HDC) is a promising approach for energy-efficient edge machine learning (ML), where low latency, low power, and tight memory budgets are essential. However, traditional HDC relies on symbolic binding and pseudo-random high-dimensional vectors, which require large dimensionality and heuristic updates to reach competitive accuracy, limiting deployment on edge hardware. We introduce XL-HD, a deterministic, projection-based, fully learnable HDC framework tailored for in-memory acceleration within edge computing systems. The method uses a fixed Sobol sequence to project binary inputs, extending learning beyond conventional HDC. During training, class prototypes are optimized in real-valued space and later binarized, enabling an entirely binary dot-product inference pipeline ideal for IMC hardware such as ReRAM crossbars. XL-HD achieves competitive accuracy on MNIST, UCIHAR, and ISOLET while maintaining a compact IMC-based inference engine with $0.395 \ \text{mm}^2$ area and only $0.40 \ μ\text{J}$ per single-cycle inference.

Problem

• Hyperdimensional Computing
• Edge Machine Learning
• In-Memory Acceleration
• Deterministic Projections
• Binary Inference

Innovation

• Hyperdimensional Computing
• Deterministic Projection
• In-Memory Computing
• Learnable HDC
• Binary Inference