Faster Convolution Inference Through Using Pre-Calculated Lookup Tables

TL;DR

This paper proposes a method to accelerate convolution inference by using pre-calculated lookup tables for low-cardinality activations, enabling faster computation and simpler hardware design.

Contribution

It introduces a lookup table-based algorithm for CNN inference that offers performance improvements and hardware simplification over traditional methods.

Findings

Significant speed-up in convolution inference using lookup tables

Simpler CNN hardware architecture possible with this approach

Potential for extending the algorithm beyond current capabilities

Abstract

Low-cardinality activations permit an algorithm based on fetching the inference values from pre-calculated lookup tables instead of calculating them every time. This algorithm can have extensions, some of which offer abilities beyond those of the currently used algorithms. It also allows for a simpler and more effective CNN-specialized hardware.