Deep learning models face the threat of adversarial attacks, which challenges their application. Moving Target Defense (MTD) is a defense paradigm that thwarts attacks by constantly changing the targets’ features and restricting the predictability of targets. Recent works have applied MTD in adversarial defense but rely on maintaining a model set and assuming a weak adversary, which causes extra storage and unreliable evaluation of the robustness of the methods. This paper proposes the MTD-Net that realizes MTD in a single neural network. In the training stage, MTD-Net parameters are randomly disabled to ensure desirable accuracy on diversified parameter groups. During each query, MTD-Net dynamically chooses several groups of parameters and aggregates their inference results for final prediction. The parameters MTD-Net chooses for inference are unpredictable, even for adversaries possessing the model’s weights. Thus, MTD-Net achieves factual unpredictability under strong whitebox attacks. We evaluate MTD-Net on two widely used datasets, i.e., GTSRB and CIFAR10. The experimental results demonstrate that MTD-Net achieves superior performance compared to existing MTD defense under adversarial attacks and is applicable to multiple architectures.

Secret image sharing (SIS) can divide a secret image into several shadow images for protection. Information hiding in the sharing domain (IHSD) fuses SIS and information hiding (IH) to simultaneously share any secret image and hide any information, and this technique can be applied in cloud computing, law enforcement and medical diagnoses. IHSD not only marks shadow images with information to prevent malicious tampering and for convenient management, search and identification but also enhances the robustness of IH. In this paper, we first introduce a formal definition of IHSD. Then, we describe a general IHSD model and algorithms with a concrete example in detail. In IHSD, we design the random element utilization model to control the random pixels generated from SIS. Then, we obtain shadow images with hidden information to realize SIS and IH simultaneously. The inputs of SIS with secret images, steganography and extra information in algorithms are without any limitations. Theoretical analyses, experiments and comparisons are presented to prove the effectiveness and feasibility of IHSD. • IHSD is short for information hiding in the sharing domain. • IHSD fuses secret image sharing and information hiding for simultaneous realization. • IHSD is realized by the random element utilization model to control randomness. • IHSD is a general idea applicable to any secret image and any information.

In some particular scenes, the shadows need to be given different weights to represent the participants’ status or importance. And during the reconstruction, participants with different weights obtain various quality reconstructed images. However, the existing schemes based on visual secret sharing (VSS) and the Chinese remainder theorem (CRT) have some disadvantages. In this paper, we propose a weighted polynomial-based SIS scheme in the field of GF (257). We use k , k threshold polynomial-based secret image sharing (SIS) to generate k shares and assign them corresponding weights. Then, the remaining n − k shares are randomly filled with invalid value 0 or 255. When the threshold is satisfied, the number and weight of share can affect the reconstructed image’s quality. Our proposed scheme has the property of lossless recovery. And the average light transmission of shares in our scheme is identical. Experiments and theoretical analysis show that the proposed scheme is practical and feasible. Besides, the quality of the reconstructed image is consistent with the theoretical derivation.