DeepAuth: A DNN Authentication Framework by Model-Unique and Fragile Signature Embedding

Authors: Yingjie Lao, Weijie Zhao, Peng Yang, Ping Li9595-9603

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

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive evaluations on various models over a wide range of datasets demonstrate the effectiveness and efficiency of the proposed Deep Auth.
Researcher Affiliation Collaboration 1 Department of Electrical and Computer Engineering, Clemson University, Clemson, SC 29634, USA 2 Cognitive Computing Lab, Baidu Research, Bellevue, WA 98004, USA
Pseudocode Yes Algorithm 1: Key Sample Generation; Algorithm 2: Signature Embedding
Open Source Code No The paper does not provide concrete access to source code (no specific repository link, explicit code release statement, or code in supplementary materials) for the methodology described in this paper.
Open Datasets Yes We leverage Le Net5 for MNIST, VGG16 for CIFAR10, and Res Net50 for Image Net.
Dataset Splits Yes We assume the trusted party has access to a held-out validation dataset for evaluating the performance. We leverage Le Net5 for MNIST, VGG16 for CIFAR10, and Res Net50 for Image Net.
Hardware Specification No The paper does not provide specific hardware details (exact GPU/CPU models, processor types with speeds, memory amounts, or detailed computer specifications) used for running its experiments.
Software Dependencies No Our implementation was based on the Paddle Paddle deep learning platform. No specific version numbers for software dependencies are provided.
Experiment Setup Yes We use 0.0005, 0.002, 0.02 for MNIST, CIFAR10, and Image Net, respectively. The learning rate (Lr) of the embedding process is given as 1e-4, 1e-3 for MNIST and 1e-6, 1e-5 for CIFAR10. For fine-tuning, the learning rate is 1e-4.