Virtual Intelligence: A Systematic Review of the Development of Neural Networks in Brain Simulation Units

The functioning of the brain has been a complex and enigmatic phenomenon. From the first approaches made by Descartes about this organism as the vehicle of the mind to contemporary studies that consider the brain as an organism with emergent activities of primary and higher order, this organism has been the object of continuous exploration. It has been possible to develop a more profound study of brain functions through imaging techniques, the implementation of digital platforms or simulators through different programming languages and the use of multiple processors to emulate the speed at which synaptic processes are executed in the brain. The use of various computational architectures raises innumerable questions about the possible scope of disciplines such as computational neurosciences in the study of the brain and the possibility of deep knowledge into different devices with the support that information technology (IT) brings. One of the main interests of cognitive science is the opportunity to develop human intelligence in a system or mechanism. This paper takes the principal articles of three databases oriented to computational sciences (EbscoHost Web, IEEE Xplore and Compendex Engineering Village) to understand the current objectives of neural networks in studying the brain. The possible use of this kind of technology is to develop artificial intelligence (AI) systems that can replicate more complex human brain tasks (such as those involving consciousness). The results show the principal findings in research and topics in developing studies about neural networks in computational neurosciences. One of the principal developments is the use of neural networks as the basis of much computational architecture using multiple techniques such as computational neuromorphic chips, MRI images and brain–computer interfaces (BCI) to enhance the capacity to simulate brain activities. This article aims to review and analyze those studies carried out on the development of different computational architectures that focus on affecting various brain activities through neural networks. The aim is to determine the orientation and the main lines of research on this topic and work in routes that allow interdisciplinary collaboration.


Introduction
Brain function studies have been highly relevant to different scientific disciplines for a long time. The sciences have focused on recognizing those epistemological aspects of brain function (how is knowledge generated?).
Throughout history, different thinkers have been interested in understanding how the brain works. Descartes offered the first approaches of relevance for these studies. A separation between the body and the mind was proposed, thus originating a dualistic problem in which cerebral activities belong to a world outside the material or human world. This approach to the mind at a purely ideological level is housed in theoretical assumptions about brain functioning since the mind is an immaterial entity without the possibility of experimentation "only through the subjective exercise of the knowledge: here and now", Descartes' vision brought an endless discussion of how the brain works, resulting in the classic "mind-brain problem". This approach and the difficulty of avoiding of brain measurement [13][14][15], which show the participation of certain brain parts in these activities or processes, the explanation of the functioning of consciousness is not simple. Different aspects such as the biochemistry presented during conscious activity or the generation of meta-representations about lived experiences [16] still represent an enigma that requires greater depth for computational studies in neuroscience.
However, this does not mean they should only reduce the explanation of consciousness or mental activities to their biological roots. Scientific advances show the possibility of exploring this enigmatic activity with new techniques and technologies that would make it possible to decode a vital part of this phenomenon [17,18]. One of the aims is to elaborate more specific explanations of the operation of this activity or emergent property of human life.
Many complex brain activities are derived from conscious experiences and the correlation between a neural system and its environment. These conscious activities, which occur in the subjects, are verifiable through the verbal manifestation of the issue experiencing these states and the detection of neuronal activity and plasticity through the synaptic exchanges derived from the action potentials executed [19] during a conscious brain process [20,21]. Conscious life on a biological level is presented through the stimuli generated from a subjective experience. These stimuli are defined through neural patterns of activation that determine the nature of the neural representation [22]. In turn, a neural model is translated as a code or language, which the brain uses to represent the information of the stimuli [23]. The task of analyzing this code is complex and can currently find its conceptualization within the quantitative descriptions offered by the nerve conduction equations of the computational models of experimental neuroscience [24,25].
Cognitive science development has brought with it the analysis and application of many faculties of the mind to mechanisms or devices [26]. Faculties such as vision, language or memory have been developed in "intelligent machines" [27,28]; however, these types of faculties are considered at a minor or primary level to the more complex activities of human nature. The development of these mental faculties in different machines has resulted from elaborating execution codes within computational languages. Their explanation finds meaning within them [29]. Currently, there are a series of questions about the possibilities that technology has to assimilate or become aware of an automaton mechanism [30].
Both systems (the mental and the computational) "process information" from their environment. Therefore, they are considered equivalent cognitive systems [18,31,32], assuming that two processing systems similarly execute certain tasks in such a way that they cannot be distinguished from one another; these could cause them to be considered the same or identical [33]. The task is focused on recognizing not the operations' similarity between one system and another, but instead, the information method processed [34].
Computational models of the mind have developed an architecture in which they work by simulating the different processes controlled by different zones or areas of the brain, trying to reproduce these processes in the operation of a processor [35,36]. These mental processes synchronize and prioritize human activities working individually or in sync with other areas [37] which are found in some brain regions and act together in human activities. Most of the architectures in brain functions focus their attention on the primary elements of the brain. They have developed promising models, as in the case of neural networks [38][39][40].
What kind of digital architectures can be effective models for this task? The problem focuses on the difference in information processing and interpretation capabilities between an individual and a system.
Regarding the study of high-order brain activities or conscious human activities, a series of projects worldwide have focused on developing simulators and platforms that aim to execute certain brain activities. Currently, some models seek to simulate brain functions in real time and speed in synaptic processes, data management and the capacity to amalgamate certain brain activities related to consciousness, such as decision making.
Understanding the functioning of consciousness has become essential for neuroscientific research today. Therefore, it is necessary to consider multidisciplinary work as the basis of any approach to this organism (since we can understand biological behaviour). However, the subjective and phenomenological factors that make each individual a unique being continue to be incomprehensible to scientific understanding. This paper aims to conduct a systematic review of different papers, their lines of action and their methods focused on studying brain function within computational neurosciences. The purpose is to recognize the main works on the development of brain simulators based on neural network architecture and to identify those aspects that can be studied in more detail in this line of research. The paper is organized as follows: Section 2 describes the construction of virtual intelligence: a systematic review of the development of neural networks in brain stimulation units and the performance metrics used to evaluate the model and presents the results of the different experiments, while Section 3 offers some discussion. Finally, Section 4 summarises the conclusions drawn from this work.

Methods and Results
The systematic review of the literature shows the results obtained in the search and classification of different papers aimed at developing this computational architecture used to stimulate brain function.
The present study found three main lines of research for the development of neural networks:
Brain plasticity and learning.

3.
Processing and execution of cognitive tasks in real real time.
Some research has taken advantage of the development of imaging studies to gain insight into biological functioning and the different brain areas involved in cognitive processes. It has provided an excellent opportunity to develop computational architectures that aim to mimic the brain as precisely as possible within scientific studies in AI.
The investigations within the computational neurosciences have developed complex structures seeking to reconstruct the biological, anatomical and cognitive processes. They are present in this brain activity by developing diverse computational architectures. It interacts with their actions (neuromorphic computing, silicon chips, neural networks, multiprocessors and others).

Research Question and Objectives
At first, the establishment of the PICO question was proposed, which could offer a search route within this investigation to make it possible to trace the corresponding work methodology. Therefore, to carry out this systematic review, we started with the following research question: What are the applications of neural networks in computational architectures for the study of brain functions?
At the time of identifying the object of study, the search terms and criteria used to answer the research question were obtained as follows, as shown in Figure 1: A. Identify the primary research into developing neural networks for brain stimulation of conscious activities. B. Determine the key concepts within the research question for the article search. C. Determine the exclusion or inclusion factors for the article search.

Inclusion and Exclusion Criteria
We carried out the review and analysis of three specialized databases in engineering and technology. The databases consulted are the following: Once the question was established, we focused on designing a search strategy to locate the most relevant information for this review. Specific exclusion criteria were needed to create the search strategy, such as the search criteria and the literary resources. It began by using the word neural network and awareness or brain as connectors as a criterion for search engines. From this, all academic articles written between 2017 and 2021 (full text) in English were selected, oriented to topics such as the brain, cognition, neural networks, learning, neurons, brain-computer interfaces, data analysis, machine learning, intelligence artificial, deep learning, meta-analysis, behaviour, neuroscience, communication and technology.
As shown in Table 1, a total of 3328 documents were found, organized as follows:

Data Extraction
The selection criteria for our research question were provided with greater clarity to identify relevant studies. The answer to the research question, as well as more specific exclusion criteria, took as criteria for the selection of publication topics oriented to research methodology, research, research funding, machine learning, brain, brain computational interfaces, artificial intelligence, cognition, meta-analysis, cognitive training, algorithms, data analysis, decision making, physiology brain, cognitive ability, deep learning and artificial neural network as shown in Table 2. Finally, relevant articles through criteria, such as reviewing titles, summaries, or full texts of the works were selected. Studies focused on treatments or analysis of the behaviour of patients with diseases, brain medical studies, or anatomical research through neuro-imaging that were not oriented to the development of computational architectures were discarded.
In addition, we considered three brain study areas as reference points for computational architectures oriented to the simulation of higher-order brain processes. The areas taken as a decision pattern for the selection of the papers are highly relevant for studies in computational neuroscience because it is through them that the different architectures addressed in this study are implemented. The areas we refer to are the following: plasticity, memory and learning.
As a result, we obtained 21 academic articles that met the characteristics of the selection criteria.
The next step was to complete the reading of the selected works and extracting and synthesizing the recovered data. We used the quality assessment results to guide the interpretation of the review's findings to indicate the types of computational architectures developed between 2017 and 2021 geared towards executing tasks similar to those of the human brain.
The data results are as follows: Identify journals that published the papers, discuss the main lines of research in neural networks to the development of brain simulation units, report and discuss review findings according to the research questions and group them into areas. During the discussion, we interpreted the review results within the context of the research questions, in a broader context related to the current international interest in the brain. We provided some related works to support the findings.

Statistical Analysis
Initially, we grouped the information obtained by the type of journal and the number of publications collected. The exposure of information was carried out through the segmentation by percentages found in each of the types of journals indicated as show in Table 3.
We observed that 28.56% of the publications were found in 3 of the 21 reviewed journals (IEEE Transactions on Biomedical Engineering, IEEE Transactions on Neural Networks and Learning Systems and Frontiers in Neurorobotics with 9.52%, respectively); also, 71.44% of the remaining articles were distributed in one publication each for the 252 journals analysed (15 reports corresponding to 4.76% for each publication). The neuroscience of spatial navigation and the relationship to artificial intelligence Machine learning to analyze brain datasets (work of brain functions, measure of brain activities)

Frontiers in Neurorobotics 2
From near-optimal Bayesian integration to neuromorphic hardware: a neural network model of multisensory integration-a brain-inspired model of theory of mind The development of different computational architectures with the use of neural networks has the task of mimicking brain functions as precisely as possible. Having the rational capacity that humans show has been one of the main objectives for studies in computational neuroscience. Knowledge of forms of learning has its primary basis in the information inputs that come from the experience of individuals with the environment. To acquire this knowledge, they need input channels through which the information reaches the brain. The main input channels for humans are the senses, which receive information and transform it into input data of different characteristics.
As shown in Table 4, the input channels found in the selected papers are the following: Table 4. Types of inputs found in the selected papers.

Input Data Paper Related
Inputs data through visual datasets or experience in real time 12 Inputs data through language datasets or experience in real time 2 Others (mixes inputs data and neural connection) 5 The input data that predominate in the development of computational architectures for the simulation of brain functions come from visual stimuli, with 57.12% of the publications. In second place are mixed architectures, which take input data from the environment (visual or language [hearing or speech]) and use principles of neural connection, such as plasticity or neuronal synapses, with 23.8%, while the input data collected from language (hearing or speech) are in last place with 9.52%. The analysis found only two publications of theoretical research unrelated to experimentation, comprising 9.52%.
The data show that the primary input data for developing computational architectures for the simulation of brain functions are derived from visual information. This tendency is due to the ability to translate data from visual stimuli into machine language.
Computational neurosciences reference the senses as input channels through which the information reaches a processing centre. However, it has not been possible to imitate the senses for the transcription of data that AI can transform into knowledge.
The review found that the use of neural networks to develop computational architectures is oriented toward the design of the networks, followed by learning algorithms to simulate different brain functions in 38.1% [41][42][43][44][45][46][47][48]. Next, the development of brain simulation software is 14.3% [49][50][51], and the development of hybrid architectures (using brain computing interfaces supported by neuromorphic processors) accounts for 14.3% [52][53][54]. The development and improvement of brain computing interfaces was 9.5% [55,56], as was analysis and database storage through machine learning [57,58]. Finally, 4.8% was shown for the design of neuromorphic processors [59]. The remaining 9.5% of the selected publications are critical reviews on the development of AI in brain stimulation through theoretical models [60,61].
The computational architectures developed in the reviewed publications are shown in Table 5: Finally, we found four application areas in which neural networks work to develop computational architectures oriented to the simulation of brain functions. These areas or units were classified as follows: • New generation data collection for the study of brain activities. • Reproduction of the brain structure and its primary functions through the development of brain simulation software to study molecular and subcellular aspects and neuronal functioning at a micro-and macroscopic level is shown. • Development of cognitive computing through neuromorphic processors and silicon chips. • The design of brain models for executing brain tasks, such as behaviour, decision making and learning.

Database Publication
Design of neural networks and learning algorithms for classification, predictions, memory, and learning. 8 Brain software simulation 3 Hybrid architectures 3 Machine learning to analyze brain datasets (work of brain functions, measure of brain activities) 2 Brain-computer interfaces 2 Development of neuromorphic processors (information processing, neural connectivity in real time [brain synapse], and learning) 1 A critical review of the development of AI in brain stimulation through theoretical models 2 The parameter used to establish paper selection measures the degree of relevance on a scale of 0 to 10, where 0 is considered unrelated, and 10 is regarded as a complete relationship. Figure 2 shows the results obtained from the analysis carried out regarding the relationship of the papers with the degree of relevance in each area.

Discussion
This systematic review shows the main studies that dealt with the development of computational architectures designed to simulate brain function between 2017 and 2021, as shown in Figure 3. We observed that the principal applications of these studies are directly related to developing cognitive processes.
The architectures found are strictly logical which considers the possibility of structuring architectures that may be capable of simulating some of the most complex human brain activities or activities related to consciousness.
One of the essential aspects of our review was the low prevalence of studies in this area in the 3328 publications we searched. Only 21 of them studied this phenomenon which means that only 0.63% of the comprehensive studies on neural networks were directed to this research area, according to the data presented (2.75% of the 762 studies found with specific criteria). It should be noted that most of the publications were directed to the development of the design of neural network structures for medical use (learning through databases for detection of diseases such as cancer, Parkinson's, Alzheimer's, among others). One of the principal challenges found in this study is the complexity of the multiple data inputs of the human being and how these inputs are translated into information for the brain. Today, the main data inputs for this type of architecture come mainly from visual or language stimuli that can be translated through binary language. However, there is still much to be done in the case of other data inputs for the human (touch, taste, smell) that allow a better understanding of the environment and, therefore, better performance.
Furthermore, one of the main challenges of computational neuroscience is to design components or architectures based on the human structure and its functioning. Currently, the construction of computational elements with characteristics that can more accurately simulate brain processes is proposed, from basic biological functions to complex brain functions (development of sensors and sensory infrastructure, neural networks with the ability to simulate chemical and electrical exchanges, synaptic capacity, sensory motor strategies for control and cognition and knowledge transfer and diffusion), all of this responding to the initial question of our research: how is knowledge generated?
This systematic review found specific lines of research in recent works and publications as show in Figure 4. The application of neural networks is the main structure for generating computational architectures for the development and simulation of brain functions. These works mainly focused on the application of computational architectures for the study of the following functions: A. Memory. B. Learning. C. Decision making or free will. We observed that the main ways for the collection, handling and storage of input data and for the implementation and development of computational architectures based on neural networks are the following: • Recognition, classification, memory and data deduction through inputs through visuals. • Acoustic or speech channels that are processed through electrical action potentials. • Hybrid architectures that have neuromorphic processors with better capacities and speeds. • Algorithms and new forms of synaptic exchanges based on chemical processes.
Faced with these challenges, international projects and initiatives have been created that aim to generate a brain and neuroscientific research network that contributes and disseminates the results obtained from its scientific practice. All this is from a multidisciplinary perspective using top-level techniques that allow the observation and measurement of phenomena related to the brain and its functioning.
Within these projects, lines of research were identified that study brain functions and the areas of the brain that participate in these processes through different investigations. The principal research projects focus on the collection, management and dissemination of data at an international level in neuroscience, brain simulation, theoretical approaches and multiscale models of the brain, brain computing focused on data analysis and robotics and imaging techniques for brain mapping and observation of brain processes.
Among the main initiatives, we can point out the following: The objective of these initiatives is to design new methods and technologies for the study of the human brain, promote new models and images for brain exploration by studying the behaviour of individual cells and the interaction with neural circuits in time and space, create and operate scientific research structures for brain research, cognitive neuroscience and other scientific disciplines related to neuroscientific studies. The results of their work are available through their websites through different resources: journal publications, scientific achievements and deliverables by project phase.
The results obtained from our work and the analysis of recently published literature have shown that current works focus on developing one or more joint applications for the development of computational architectures. In addition, it should be noted that we identified six lines of research that verify these findings during our systematic review. Within the lines of research for the area of computational neuroscience, the following stand out: A. Large-scale recording and modulation in the nervous system. B. Next generation brain imaging. C. Integrated approaches to brain circuit analysis. D. Neuromorphic computing with biological neural networks as analog or digital copies in neurological circuits. E. Brain modeling and simulation.
These findings can agree with the results obtained in this systematic review on the main applications of neural networks in the development of computational architectures ( Figure 5).
It is worth mentioning that the projects and initiatives mentioned are investigations that work in medium and long-term objectives and show their results through publications within their platforms where the main advances in their different research areas can observe in their platforms. These initiatives stand out among the most important projects around the world for their vision and multidisciplinary and collaborative work with international organizations such as universities, health systems and research centers.

Conclusions
The use and application of neural networks in the development of computational architectures for the execution of tasks that involve the human brain is undoubtedly a growing field of research. Currently, computational neuroscience studies have managed to imitate functions of a rational nature for the human mind. However, multiple factors still keep more complex brain tasks at a distance.
Advances in computational neuroscience studies have proven very useful in understanding new unknowns in brain behaviour. Consequently, multiple types of computational architectures are applying this new knowledge in the different areas of IT.
Applying different technologies is a fundamental step in studies of the brain and knowledge of its functioning. This knowledge allows the development of many types of computational architectures based on a structure close to that of brain function. Neural networks have made it possible to simulate many logical processes of the brain and how a neuron can behave in order to develop devices with a specific capacity or type of intelligence.
As mentioned in the previous section,different initiatives are working to discover new pathways towards understanding the brain. It is extremely important to delve into the work from these initiatives after this study, since many of their works are still under investigation.
Although advances in neuroscience have made it possible to delve into the functioning of brain processes at a biological and chemical level, human behaviour cannot be the object of biological or physical reductionism in this area. However, scientific advances have shown that achieving a complex and holistic understanding of brain function requires a multidisciplinary vision, which must start from the basic physical foundations of this process and be supported by emerging elements that technology could offer.
Knowledge of brain behaviour is necessary for understanding human behaviour, but this type of knowledge can also be of vital importance for issues related to the prevention and treatment of brain diseases,.
Finally, the purpose of examining each of these approaches is the need to offer a general theoretical vision of how studies in this area have been investigated and to point out their main discussions. It should be noted that the approaches offered are a way of showing what the main objectives around this topic have been. Within the main tasks for the study and development of research in this field of computational neuroscience we can point out: • The need to build multidisciplinary projects for the study of brain functions. • The application of multiple technologies for the simulation and modeling of brain activities. • The creation of a global platform of free access for researchers.
However, computational neuroscience studies have as their main challenge one of the most enigmatic parts of the brain, consciousness. Knowledge of the brain functions that involve consciousness and the different areas of the brain that participate in this process is one of the main objectives of these disciplines. The possibility of developing devices capable of recognizing themselves and interacting with their environment is one of the most critical challenges for research in these disciplines and for future developments in AI.