Probabilistic Selective Encryption of Convolutional Neural Networks for Hierarchical Services

Jing-zhen, Tian; Zhou, Jiantao; Duan, Jia

doi:10.1109/cvpr46437.2021.00224

Cited by 9 publications

(8 citation statements)

References 23 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Tian et al [16] proposed PSS to select the important parameters from the convolutional layer parameters of CNN. Many studies [24][25][26][27] have shown that the parameters in CNNs are not of equal importance, so the performance of CNNs can be controlled by selecting a small number of more important parameters.…”

Section: Probabilistic Selection Strategymentioning

confidence: 99%

“…The number of layers where important parameters are located in VGG19 The reasons for choosing the VGG19 model are as follows. First, according to the study of Tian [16], encrypting the parameters of the VGG model by a few layers can control the model's performance to a greater extent. Second, VGG19 is a classification model with a greater tolerance for fluctuations in the output results in terms of values than models with complex output results such as the style migration model.…”

Section: Parameter Valuementioning

confidence: 99%

“…For resource-constrained devices, encryption and decryption are very expensive, so some schemes use a selection strategy to encrypt some of the parameters in the model. Tian et al [16] propose a probabilistic selection strategy (PSS) to determine the importance of parameters in convolutional neural networks (CNN) and protect CNN by encrypting important parameters. Chen and Wu [17] propose a novel framework for deep-learning models that provides access control to the trained deep neural networks so that only authorized users can utilize them properly.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

A Neural Network Model Secret-Sharing Scheme with Multiple Weights for Progressive Recovery

et al. 2022

View full text Add to dashboard Cite

With the widespread use of deep-learning models in production environments, the value of deep-learning models has become more prominent. The key issues are the rights of the model trainers and the security of the specific scenarios using the models. In the commercial domain, consumers pay different fees and have access to different levels of services. Therefore, dividing the model into several shadow models with multiple weights is necessary. When holders want to use the model, they can recover the model whose performance corresponds to the number and weights of the collected shadow models so that access to the model can be controlled progressively, i.e., progressive recovery is significant. This paper proposes a neural network model secret sharing scheme (NNSS) with multiple weights for progressive recovery. The scheme uses Shamir’s polynomial to control model parameters’ sharing and embedding phase, which in turn enables hierarchical performance control in the secret model recovery phase. First, the important model parameters are extracted. Then, effective shadow parameters are assigned based on the holders’ weights in the sharing phase, and t shadow models are generated. The holders can obtain a sufficient number of shadow parameters for recovering the secret parameters with a certain probability during the recovery phase. As the number of shadow models obtained increases, the probability becomes larger, while the performance of the extracted models is related to the participants’ weights in the recovery phase. The probability is proportional to the number and weights of the shadow models obtained in the recovery phase, and the probability of the successful recovery of the shadow parameters is 1 when all t shadow models are obtained, i.e., the performance of the reconstruction model can reach the performance of the secret model. A series of experiments conducted on VGG19 verify the effectiveness of the scheme.

show abstract

Section: Probabilistic Selection Strategymentioning

confidence: 99%

Section: Parameter Valuementioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

A Neural Network Model Secret-Sharing Scheme with Multiple Weights for Progressive Recovery

et al. 2022

View full text Add to dashboard Cite

show abstract

“…The model owner can only prove his or her copyright by verifying the watermark after the model is stolen. Therefore, some researchers have proposed the active authorization control method for models 6‐14 . If users can prove their legitimate right to use the model, they can activate the reasoning function of the model.…”

Section: Introductionmentioning

confidence: 99%

“…They added a control layer as the last layer in the DNN model. The control layer of the DNN model is deleted when entering a user's fingerprint, and then the reasoning function of the DNN model is used normally.Tian et al12 encrypted important parameters in the DNN model with a selective encryption algorithm. Different numbers of parameters are decrypted for different users, and hierarchical access services are provided to users.…”

mentioning

confidence: 99%

A framework for deep neural network multiuser authorization based on channel pruning

Wang

Song

Zhu

et al. 2023

Concurrency and Computation

View full text Add to dashboard Cite

SummaryVarious deep neural network (DNN) model watermarks have been proposed by researchers to verify copyrights for deep neural networks DNN. However, most DNN watermarking methods cannot prevent attackers from stealing and using the model. Unlike many existing approaches, this paper uses a channel pruning algorithm to protect DNN models, which verifies DNN models copyrights but also prevents the illegal use of DNN models. In this work, the pruning threshold or pruning rate is used as the secret key of a DNN model. After the secret key is distributed to multiple users, they prune the DNN model with the secret key, and the pruned and fine‐tuned model is provided to the users. The users can verify ownership of the model according to the pruning accuracy and fine‐tuning accuracy. If the secret key is incorrect, the accuracy of the model after fine‐tuning will be very low, and users will be unable to use the reasoning function of the fine‐tuned model. Based on the CIFAR‐10 and CIFAR‐100 datasets, we conducted experiments on five popular DNN models. The experimental results show that we can authorize multiple users by pruning very few channels in the convolution layers of the DNN model.

show abstract

Active Authorization Control of Deep Models Using Channel Pruning

Wang

Song

Zhu

et al. 2023

Communications in Computer and Information Science

View full text Add to dashboard Cite

Probabilistic Selective Encryption of Convolutional Neural Networks for Hierarchical Services

Cited by 9 publications

References 23 publications

A Neural Network Model Secret-Sharing Scheme with Multiple Weights for Progressive Recovery

A Neural Network Model Secret-Sharing Scheme with Multiple Weights for Progressive Recovery

A framework for deep neural network multiuser authorization based on channel pruning

Active Authorization Control of Deep Models Using Channel Pruning

Contact Info

Product

Resources

About