This repository contains the reproducibility package for the manuscript:
Auditable Neural Differential Distinguishers for Reduced-Round Lightweight-Style Block Ciphers: A Reproducible Machine-Learning Framework for Cryptographic Security Evaluation
This repository provides code and data for a controlled study of machine-learning and deep-learning neural differential distinguishers on reduced-round lightweight-style block ciphers.
The educational cipher TinySPN-32 is not secure and must not be deployed. It is used only as a transparent benchmark for reproducible experiments.
- src/: source code for TinySPN-32, dataset construction, model training, evaluation, and plotting.
- data/: generated experimental data used in the manuscript.
Default experiment:
- Delta: 0x0000000F
- Rounds: 1 to 6
- Feature size: 96 bits
- Models: logistic regression, random forest, MLP, Bit-CNN
This repository does not provide an attack against any deployed or standardised cryptographic system. All experiments are restricted to reduced-round or educational primitives.
Auditable Neural Differential Distinguishers for Reduced-Round Lightweight-Style Block Ciphers: A Reproducible Machine-Learning Framework for Cryptographic Security Evaluation