Auto-Tuning Fixed-Point Precision for Efficient RISC-V AI

Created using ChatSlide

This presentation delves into optimizing machine learning computation on edge devices, addressing efficiency challenges. It introduces Packed SIMD Extension, focusing on tensorization, vectorization, and a Uniform Selector Mechanism (USM) to leverage 16-bit fixed-point arithmetic. Experimental results highlight notable reductions in instruction count, improved performance speed, and maintained accuracy. The findings underscore the importance of efficient computation in modern applications,...