IEEE Computational Intelligence Magazine - November 2020 - 13

become the specified type is strictly greater than zero. Although
probabilistic methods are designed based on the probability
theories, the results are determined with certainty and without any possible errors. Probabilistic methods were first
introduced as the main foundation of fuzzy logic by Erdos
and Spencer [6]. To evaluate the outcomes of a system based
on computational intelligence, probabilistic methods are
mostly defined by randomness [12]. Accordingly, probabilistic
methods can provide proper solutions to problems based on
prior knowledge.
C. Issues on Fighting COVID-19

Based on our observations, there are several urgent issues related to COVID-19 that must be combatted. These issues can be
categorized into five topics: tracking and predicting virus propagation (TPVP), characterization of symptoms of virus infections (CSVI), treatment design (TrD), precaution development
(PD), and public health policy making (PHPM). The issues of
each topic are listed below:
1) Tracking and Predicting Virus Propagation (TPVP)
* Surveillance and tracking of COVID-19-infected
patients.
* Modeling and predicting virus propagation and pathways.
* Visual analytics techniques and applications for propagation modeling and monitoring.
2) Characterization of Symptoms of Virus Infections (CSVI)
* Discovery of early markers/symptoms of viral infections.
* Personalized and group-based risk profiling and prediction.
* Real-time and early alerting systems for hazardous and
forefront outbreaks.
* Fast and accurate diagnosis of COVID-19 through analytics and modeling using various biomedical data, e.g.,
images, vital signs, genome, etc.
3) Treatment Design (TrD)
* Treatment optimization and care planning for the best
care of patients.
* Prognosis and outcome prediction for patients for
effective resource allocation.
* Drug discovery and repurposing through big data analytics approaches.
4) Precaution Development (PD)
* Vaccine design through machine learning approaches.
* Intelligent analysis of social media and networks for
contact tracing and safety control.
* Integrations of intelligent computing mechanisms with
information technology systems and the internet of
things (IoT) for smart care in COVID-19.
5) Public Health Policy Making (PHPM)
* Secure and privacy-preserving analysis of data in public
health emergencies.
* Public health policy making through big data analytics
and model simulations.

To overcome these issues, researchers are actively conducting research to obtain various outcomes, and many have adopted computational intelligence. Since there are so many studies
that address the issues associated with COVID-19, this survey
focuses on studies designed using computational intelligence
were selected for discussion in the following sections.
III. Neural Networks for Combating COVID-19

As mentioned earlier, an artificial neural network applies the
principle of deep learning and achieves a high level of representation learning. Representation learning is a learning method that can automatically learn representations from data.
Learning algorithms do not require humans to help them
extract features. Accordingly, neural networks can be easily utilized for extracting important representations of virus propagation and the characteristic symptoms of a viral infection, among
other factors. However, deep neural networks can extract useful
knowledge mostly if the amount of data is sufficient, which
means they can hardly deal effectively with incompleteness or,
most importantly, data with missing values in the processing
model. Fortunately, many countries have established a series of
COVID-19 data collection mechanisms. As a result, building an
effective deep neural network to fight COVID-19 is possible.
For example, many researchers have built deep neural network
models for characterizing viral infections by using CT chest
images, as shown in Fig. 1, or X-ray images [23], [24]. Table I
summarizes the issues that have been addressed by existing
neural network methods. Owing to their excellent ability to
extract important representations, most applications involving
neural networks have addressed the issue of CSVI, as shown
in Table I.
Roy et al. [14] proposed the application of artificial neural
networks for analyzing lung ultrasonography (LUS) images.

FIGURE 1 Example of computed tomography (CT) image [13].

NOVEMBER 2020 | IEEE COMPUTATIONAL INTELLIGENCE MAGAZINE

IEEE Computational Intelligence Magazine - November 2020

Table of Contents for the Digital Edition of IEEE Computational Intelligence Magazine - November 2020