Neurocomputing 525 (2023) 54–56

Contents lists available at ScienceDirect

Neurocomputing
journal homepage: www.elsevier.com/locate/neucom

Preface

Guest Editorial: Special issue on encoding-decoding-based state

estimation for neural networks

1. Introduction State estimation for Neural Networks and develop some new para-
digms for systems analysis and estimator design.
The rapid development of artificial intelligence (AI) has been In summary, state estimation for neural networks whose mea-
profoundly changing our daily life as well as the human society. surements are transmitted via networks, particularly in the case
The artificial neural network (ANN) is one of the key cornerstones of coding-decoding-based communication, have already become
of the development of AI technology, which has once again of vital importance in academia and industry. The relevant results
attracted widespread attention all over the world. Artificial neural can be used to analyze the complex dynamic behavior of neural
network, also referred to as neural network, is a mathematical or networks and are useful for the effective implementation of vari-
computational model that mimics the structure and function of ous engineering tasks.
biological especially human brain neural networks. This idea of
mimicking the function of biological brain directly affects the 2. This special issue
development of AI technology. It should be noted that it is of great
significance for the development of AI to better realize the ‘‘intel- The purpose of this special issue is to bring the latest
ligence-like-brain”, thereby completing the decision-making approaches in encoding-decoding-based state estimation for neu-
behavior, program behavior and reflection behavior of the human ral networks. The list of possible topics includes, but is not limited
brain. to: state estimation of NNs subject to network-induced phenom-
It is nowadays widely acknowledged that the neuron states play ena under CDBCMs, secure state estimation of NNs subject to
a pivotal role in the utilization of neural networks (NNs), which, cyber-attacks under CDBCMs, resilient state estimation of NNs
unfortunately, is usually unavailable in practice especially for under CDBCMs, robust state estimation of NNs under CDBCMs,
those with relatively large scales. As a consequence, the state esti- probability-constrained state estimation of NNs under CDBCMs,
mation problems for NNs have attracted considerable research state estimation/synchronization/control of complex networks
interest in both academia and industry. So far, there have been under CDBCMs, pinning control of complex networks under
plenty of research fruits available in the literature addressing the CDBCMs.
state estimation issues for NNs from different perspectives. Several In this special issue, we have solicited submissions to this spe-
techniques have been developed, some of which have already been cial issue from electrical engineers, control engineers, mathemati-
adopted in practical engineering and significantly enhanced the cians, and computer scientists. After a rigorous and careful peer-
applications of NNs. review process, the editorial team has selected 7 high-quality
On the other hand, along with the fast development of network papers from the submissions, where each paper has been reviewed
technology, many estimators have been deployed far from the sen- by at least two expert reviewers. These papers have covered both
sors, which gives rise to the remote state estimation where the col- the analysis and synthesis issues subject to various kinds of net-
lected measurements of sensors should be transmitted to work-induced phenomena under CDBCMs. Furthermore, the issue
estimators in a digital way via communication networks. In the consists of a survey paper and six research papers, which can stim-
context of digital transmission, due to the limitation of bandwidth, ulate more interests of researchers in the studies of CDBCMs so as
only finite bits of data are allowed to be transmitted at a certain to produce more theoretical and practical outcomes.
time. Therefore, coding-decoding-based communication mecha- Stability analysis plays an important role in the research for
nisms (CDBCMs) have proved to be an effective way in saving ANNs. The existence of time-delays might give rise to a significant
resources and achieving digital transmission. effect in the stability analysis of ANNs. The survey paper ‘‘A survey
It is well known that the coding-decoding-based communica- of recent advances on stability analysis, state estimation and synchro-
tion mechanism would inevitably result in errors during the pro- nization control for neural networks” by Chen et al. has provided a
cess of coding and decoding of the signals, which bring comprehensive survey on the stability analysis for different kinds
challenges and difficulties to the analysis and synthesis for remote of delayed ANNs. Furthermore, this survey paper has reviewed
state estimation issues for NNs. As such, it is of theoretical and the recent advances regarding the state estimation issues for ANNS
practical significance to investigate the coding-decoding-based subject to communication constraints induced by communication
networks (including the event-triggered mechanisms and the com-

https://doi.org/10.1016/j.neucom.2023.01.001
0925-2312/Ó 2023 Published by Elsevier B.V.
L. Ma, L. Zou, X. Yi et al. Neurocomputing 525 (2023) 54–56

munication protocol scheduling effects). Then, this survey has pro- matrix inequalities, the explicit filtering parameters have been
vided an overview on the synchronization problem for master- derived.
slave neural networks. As it is well known, many important tasks (such as state-feed-
Event-triggered mechanism is an effective way to reduce the back control, fault diagnosis) are implemented based on the infor-
communication consumption while guaranteeing the desired sys- mation of the system states, which are always obtained by using
tem performance (including the control performance and state the state estimation techniques. In the paper entitled ‘‘Observer-
estimation performance). The extensively studied event-triggered based PID control for actuator-saturated systems under binary encod-
mechanisms can be divided into two categories: the static event- ing scheme” by Wen et al., a proportional-integral-derivative (PID)
triggered mechanisms and dynamic event-triggered mechanisms, controller has been designed for a type of actuator-saturated sys-
where the former ones are implemented based on certain fixed tems affected by multiplicative noises under the binary encoding
triggering parameters, and the later ones are implemented based scheme (BES), where the PID controller is developed based on
on certain variable triggering parameters. In the paper entitled the state estimates derived by an observer. The so-called BES is uti-
‘‘Adaptive event-triggered state estimation for a class of stochastic lized in the sensor-to-observer channel to transform the measure-
complex networks subject to coding-decoding schemes and missing ment signal into a binary bit string with fixed length. Considering
measurements” by Jia et al., an adaptive event-triggered mechanism the impacts of the actuator saturations, multiplicative noises, and
has been developed for the recursive state estimation problem for effects induced by the BES, sufficient conditions have been pro-
a type of nonlinear complex dynamical networks with random posed to ensure the existence of the desired observer-based PID
coupling parameter, missing measurements and coding-decod- controller. The desired PID controller parameters have been calcu-
ing-based communication scheme. The desired state estimator lated by solving a set of linear matrix inequalities. In the final paper
gain matrix has been derived by optimizing the upper-bound for entitled A zonotope-based fault detection for multirate systems with
the trace of the resultant state estimation error covariance. Fur- improved dynamical scheduling protocols, the distributed fault
thermore, this paper has also analyzed the monotonic relationship detection problem has been studied for a class of discrete-time
between the optimization index and missing probability. In the multi-rate systems over sensor networks with an improved com-
paper entitled ‘‘Dynamic event-triggered state estimation for time- munication protocol, in which both unknown-but-bounded (UBB)
delayed spatial-temporal networks under encoding-decoding scheme” noises and faults are comprised in two zonotopic sequences. A spe-
by Sun et al., by introducing an auxiliary dynamical variable in the cial quasi-try-once-discard protocol has been constructed to
event-triggered scheme, a so-called dynamic event-triggered schedule the signal transmission between the sensor nodes and
mechanism has been constructed to adjust the triggering threshold the remote filter. The multi-rate strategy has been presented to
according to triggering error (i.e. the difference between the cur- update the sampling rate of the plant and the sensors. By using
rent measurement signal and the latest triggering measurement the mathematical induction, a series of zonotopic sets have been
data). Sufficient conditions have been obtained by constructing a derived confining the resultant filtering errors. The desired filter
Lyapunov-Krasovskii functional to guarantee the exponential parameters have been calculated by minimizing the F-radii of the
mean-square ultimate boundedness of the state estimation error resultant zonotopic sets. A detection mechanism is constructed
system. In addition, the desired state estimator parameter is calcu- to examine the occurrences of faults based on the derived zono-
lated by solving certain matrix inequality. topic sets.
Cyber-attacks represent the signals that launched by adver-
saries over the communication with the purpose of deteriorating Data availability
the system performance. For the state estimation issue with
CDBCMs, the estimator should be designed with the full consider- No data was used for the research described in the article.
ation of the effects induced by the attacks. The widely studied
cyber-attacks include, but are not limited to, the denial of service Declaration of Competing Interest
(DoS) attacks, deception attacks and replay attacks. For the state
estimation issue subject to Dos attacks, the attackers would try The authors declare that they have no known competing finan-
to flood the targeted network resource such that the communica- cial interests or personal relationships that could have appeared to
tion is unavailable to the estimator. Consider the effects induced influence the work reported in this paper.
by DoS attacks, the recursive secure state estimation problem has
been studied in the paper entitled ‘‘Encoding-decoding-based Acknowledgement
finite-horizon recursive secure state estimation for dynamic coupled
networks with random coupling strength” by Meng et al., where a We would like to express our gratitude to the contributors who
sequence of variables has been utilized to depict the stochastically have submitted their high-quality papers to this Special Issue and
varying coupling strengths. To save the network resources and the anonymous reviewers who have volunteered their time to pro-
facilitate the data transmission process, an encoding-decoding vide valuable and constructive comments to improve the quality of
strategy is introduced to deal with the transmitted data. By recur- the papers submitted. We would like to sincerely thank Professor
sively dealing with two coupled backward Riccati difference equa- Zidong Wang for his constructive suggestions and help during
tions, the desired estimator parameter has been derived for the the preparation of this issue. We would also thank the Editorial
finite-horizon H-infinity state estimator. In the paper entitled Staff for their help during the whole process of this Special Issue.
‘‘Encoding–decoding-based secure filtering for neural networks under
mixed attacks” by Yi et al., both the fault injection and DoS attacks ⇑
Lifeng Ma a,
have been simultaneously considered to reflect the practical oper- Lei Zou b
ating conditions of the investigated ANNs. A set-membership filter Xiaojian Yi c
has been constructed to generate the desired state estimates for Tingwen Huang d
the time-varying ANNs under the effects of encoding-decoding a
School of Automation, Nanjing University of Science and Technology,
mechanism. Sufficient conditions have been obtained to ensure Nanjing 210094, China
that the estimation error could be confined within certain pre- b
scribed ellipsoidal region at each time instant. By solving a set of Department of Computer Science, Brunel University London, Uxbridge,
Middlesex UB8 3PH, UK

55
L. Ma, L. Zou, X. Yi et al. Neurocomputing 525 (2023) 54–56

c Received 29 December 2022

Department of Mechatronical Engineering, University of Beijing Insti-
Accepted 8 January 2023
tute of Technology, Beijing 100081, China
d
Department of Science, Texas A&M University at Qatar, Doha 23874, Available online 10 January 2023
Qatar
⇑ Corresponding author.
E-mail address: [email protected]</e

56