NeuroDeX: Unlocking Diverse Support in Decompiling Deep Neural Network Executables

TL;DR

NeuroDeX tackles the risk of DNN executable reverse engineering by introducing an LLM-assisted, dynamic-analysis pipeline that recovers operator types, attributes, and full high-level models. It systematically handles compilation optimizations, cross-architecture support, and quantized models, delivering high operator-recognition accuracy and robust model reconstruction. The approach demonstrates strong performance across TVM and GLOW backends, with quantized models achieving functional parity and reasonable inference accuracy, while maintaining efficiency advantages over prior work. The work highlights practical security implications for DL compilers and outlines directions for extending the framework to broader model families and threat scenarios. Overall, NeuroDeX provides a scalable, architecture-agnostic solution for decompiling DNN executables with significant implications for both defense and intellectual-property considerations.

Abstract

On-device deep learning models have extensive real world demands. Deep learning compilers efficiently compile models into executables for deployment on edge devices, but these executables may face the threat of reverse engineering. Previous studies have attempted to decompile DNN executables, but they face challenges in handling compilation optimizations and analyzing quantized compiled models. In this paper, we present NeuroDeX to unlock diverse support in decompiling DNN executables. NeuroDeX leverages the semantic understanding capabilities of LLMs along with dynamic analysis to accurately and efficiently perform operator type recognition, operator attribute recovery and model reconstruction. NeuroDeX can recover DNN executables into high-level models towards compilation optimizations, different architectures and quantized compiled models. We conduct experiments on 96 DNN executables across 12 common DNN models. Extensive experimental results demonstrate that NeuroDeX can decompile non-quantized executables into nearly identical high-level models. NeuroDeX can recover functionally similar high-level models for quantized executables, achieving an average top-1 accuracy of 72%. NeuroDeX offers a more comprehensive and effective solution compared to previous DNN executables decompilers.

Figures (6)

• Figure 1: BTD Cross Version Type Recognition Accuracy
• Figure 2: TSNE Dimensionality Reduction of Different Operators
• Figure 3: Workflow of NeuroDeX
• Figure 4: An Example of Disassembled Code
• Figure 5: Analysis of Quantized Compiled Models