A Quantization-Aware Training Based Lightweight Method for Neural Distinguishers

TL;DR

This paper introduces a quantization-aware training method for neural distinguishers that significantly reduces computational complexity by replacing multiplications with Boolean operations, maintaining high accuracy in cryptanalysis tasks.

Contribution

It proposes a lightweight neural distinguisher using learnable quantization, converting all multiplications into Boolean logic, thus greatly reducing complexity while preserving accuracy.

Findings

Operations reduced to 13.9% of original model

Elimination of costly 32-bit multiplications

Accuracy drops only 2.87% with quantization

Abstract

In 2019, Gohr pioneered the application of deep neural networks to differential cryptanalysis, developing DNN-based neural distinguisher classifiers to analyze the SPECK lightweight block cipher. Unlike traditional differential analysis, which relies on Boolean operations on 0-1 sequences, neural distinguishers extract continuous features, introducing 32-bit multiplications operations that increase complexity and potential redundancy. This study proposes a lightweight neural distinguisher based on quantization-aware training. Leveraging learnable step-size quantization, the model's weights are quantized to 1.58 bits, enabling the replacement of all convolutional multiplication operations with Boolean logic. Additionally, the ReLU activation function is reimplemented as a comparison-based indicator function. This transforms the original 32-bit multiplication-dependent architecture into a lightweight structure composed solely of Boolean operations, additions, and indicator functions. Experimental results confirm significant computational complexity reduction. Owing to a high proportion of zero-valued weights, the total operations amount to just 13.9% of Gohr's model. Critically, the most costly 32-bit multiplications are eliminated, with classification accuracy dropping by only 2.87%. When applied exclusively to the initial convolutional layer, the 128 1-by-1 convolutions are replaced with 4 Boolean operations on 16-bit sequences, incurring a negligible 0.3% accuracy loss.