DeepHardMark: Towards Watermarking Neural Network Hardware

Authors: Joseph Clements, Yingjie Lao4450-4458

AAAI 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate the performance of the hardware watermarking scheme on popular image classification models using various accelerator designs. Our results demonstrate that the proposed methodology effectively embeds watermarks while preserving the original functionality of the hardware architecture. Specifically, we can successfully embed watermarks into the deep learning hardware and reliably execute a Res Net Image Net classifier with an accuracy degradation of only 0.009%.
Researcher Affiliation Academia Joseph Clements, Yingjie Lao Department of Electrical and Computer Engineering, Clemson University Clemson, South Carolina, 29634 jfcleme@g.clemson.edu, ylao@clemson.edu
Pseudocode Yes Algorithm 1: Block Constrained Perturbations and Algorithm 2: Reducing Intra-block Perturbations are provided.
Open Source Code No The paper does not provide an explicit statement or link to its own open-source code for the methodology described.
Open Datasets Yes We conduct these experimental evaluations on multiple image classification models for the Cifar10, Cifar100, and Image Net datasets.
Dataset Splits No The paper mentions using a 'validation dataset' in the definition of the Fid metric, but does not specify the explicit split percentages, counts, or methodology for the training, validation, and test datasets.
Hardware Specification No The paper describes the target hardware architecture for watermarking (MMU, FPGA/ASIC design) and the software used for simulations (Pytorch), but it does not specify the hardware used to run these simulations or experiments.
Software Dependencies No The software simulations are developed using the deep learning package, Pytorch. A specific version number for PyTorch or any other software dependency is not provided.
Experiment Setup Yes Algorithm 1 lists hyperparameters: c, Tδ, TB, ϵδ, ϵB, ρ1, ρ2, ρ3. Algorithm 2 lists hyperparameter C. These parameters define the experimental setup for the algorithms.