Compiler-Driven Approximation Tuning for Hyperdimensional Computing

The authors introduce ApproxHDC, a framework that automates the identification and application of domain-specific approximations in Hyperdimensional Computing (HDC) workloads. This system extends the HPVM-HDC compiler infrastructure to enable retargetable compilation across diverse hardware backends, including CPUs, GPUs, and simulated ReRAM and PCM accelerators.

• ApproxHDC navigates an exponentially large space of possible approximations using efficient search and analysis to identify high-impact configurations.
• The framework spans both software and hardware levels to optimize performance while maintaining accuracy.
• HDC algorithms are inherently tolerant to noise and approximation, allowing for substantial performance gains with minimal accuracy loss.
• The approach addresses the physical and economic limits of Moore's law by leveraging domain-specific approaches for machine learning acceleration.

ApproxHDC enables significant performance improvements on heterogeneous hardware platforms by automatically tuning approximations that would otherwise be difficult to identify manually.