IEEE Computational Intelligence Magazine - November 2021 - 70

Frontier
Research
Zichi Wang, Guorui Feng, Hanzhou Wu, and Xinpeng Zhang
School of Communication and Information Engineering,
Shanghai University, Shanghai, 200444, CHINA
Data Hiding in Neural Networks for Multiple Receivers
Abstract
R
ecently, neural networks have
become a promising architecture
for some intelligent
tasks. In
addition to conventional tasks such as
classification, neural networks can be
used for data hiding. This paper proposes
a data hiding scheme to transmit different
data to multiple receivers via the same
neural network simultaneously. Additional
data are embedded into the neural network
during the process of training
instead of modifying the network parameters
after training. As a result, the reduction
in detection accuracy of the neural
network is trivial with additional data
embedded. On the receiver side, additional
data can be extracted using specific data
decoding networks. Instead of training,
the parameters of data decoding networks
are generated using the embedding keys
possessed by the receivers. Therefore, it is
unnecessary to store and transmit the
decoding networks secretly, and the practicability
is satisfactory. A receiver can
extract corresponding additional data
with the correct embedding key. For
other parts of additional data, the receiver
cannot affirm their existence and therefore
cannot extract them. Experimental
results verify the effectiveness of the proposed
data hiding scheme, including
embedding capacity and robustness.
Corresponding Author: Xinpeng Zhang (xzhang@shu
.edu.cn).
Digital Object Identifier 10.1109/MCI.2021.3108305
Date of current version: 13 October 2021
FIGURE 1 Different types of neural networks.
I. Introduction
In recent years, deep learning has become
the most effective technique in the fields
of artificial intelligence, e.g., image recognition
[1], semantic image segmentation
[2], and natural language processing [3].
As shown in Fig. 1, with a large scale of
labeled data and computational resources,
various network models with strong
learning capability, e.g., CNNs (convolutional
neural networks) [4], RNNs
(recurrent neural networks) [5], GANs
(generative adversarial networks) [6],
ResNets (residual networks) [7], and
CapsNets (capsule networks) [8], have
been designed for different tasks.
Except for the intelligent tasks, these
network models can be used for data
hiding which aims to transmit additional
data via digital media without causing
serious distortion on the cover media
[9]. In [10], additional data were embedded
into model parameters using a
parameter regularizer. Since the data are
embedded during the training process,
the model performance is not impaired.
Similarly, in [11], a neural network
model was enabled to learn the given
ResNets
CNN
RNN
Computation
GAN
CapsNets
70 IEEE COMPUTATIONAL INTELLIGENCE MAGAZINE | NOVEMBER 2021
1556-603X/21©2021IEEE
IEEE Computational Intelligence Magazine - November 2021

Table of Contents for the Digital Edition of IEEE Computational Intelligence Magazine - November 2021
