Adaptive resonance as a component of the mechanism of formation and development of the cognitive space of the individual in the cognitive process

Размещено на http://www.allbest.ru/

Kharkiv National Automobile and Highway University

V.N. Karazin Kharkiv National University

O.M. Beketov National University of Urban Economy in Kharkiv

Adaptive resonance as a component of the mechanism of formation and development of the cognitive space of the individual in the cognitive process

Emelyanova T.V., Yarkho T.O., Legeyda A.V., Legeyda D.V.

Within the framework of the neurodynamic concept of the activity of the cognitive space, a study of the nature and character of the mechanism of adaptive resonance and its connections with the internal mechanisms of cognitive activity was conducted. Research from the standpoint of modern cognitive science on modeling the activity of the neural system within the framework of the concept of neurodynamic organization and functional modes of the cognitive space of memory, the role of adaptive resonance of neural networks in the formation and development of the cognitive space of an individual allowed to determine the connection of adaptive resonance with the mechanisms of formation and development cognitive space of the individual.

The internal mechanisms of transformation of model representations of mental images, such as coactivation of connections in neural ensembles, concatenation of metastable states of model representations and adaptive resonance of neural activity, were investigated. The authors come to the conclusion about the close interdependence of such dynamic processes, since their relationship can be determined by the components of mechanisms and characteristics of cognitive processes.

It can be expected that the understanding of the connection of adaptive resonance with the internal mechanisms of the formation of the cognitive space will determine the direction of further research and understanding of the architecture of the cognitive process of thinking, will provide additional arguments in favor of understanding the learning process as a process of developing the cognitive space and activating the thinking of the individual. A similar understanding of the learning process at technical universities will ensure further improvement of the technologies of professional and cultural development of the individual.

The cognitive space of an individual is the space of cognitive abilities, thinking, memory, etc. In order to be able to purposefully expand the cognitive space of an individual, one should understand the mechanisms of formation and development of the basic components of cognitive abilities and the cognitive space of memory. The cognitive space of an individual is the space of cognitive functions that can be understood and studied only as a result of researching the mechanisms of cognitive activity of the brain. These studies are based on the assumption that the space of cognitive functions as the space of a dynamic nonlinear system can be studied on the basis of analysis of non-stationary processes and metastable states.

At the same time, examples of biological architectures of neural networks are discussed, with the help of which "mind is formed", "adaptive intelligence".[1]. The modeling of such architectures is based on the principles of the theory of adaptive resonance (ART) of neural networks, which is the fundamental theory of artificial intelligence at the present stage. Adaptive resonance of neural networks was considered by the authors of the theory of adaptive resonance, Steven Grossberg and Gail Carpenter [2], as the process of creating a neural network in which the previous or "old" information does not disappear with the preservation of new information. During the "training" of artificial intelligence created within the framework of old models of neural networks, the plasticity and adaptability of previously received information was not ensured.

In the paradigm of neurodynamic systems, the description of the cognitive memory space is carried out in the system of dynamic variables (the phase space of the dynamic problem), so that the model representations of mental images are integrated metastable structures in the phase space of the dynamic problem. Accepting the understanding of the stable state of the dynamic mode of neural activity as the "substrate equivalent of a psychological image", the adaptive resonance of neural networks should be considered as a guide to the constituent mechanisms of the creation of metastable structures of new mental images in the neurodynamic system.

In the article, within the framework of the neurodynamic concept of the activity of the cognitive space, the nature and the features of the mechanism of adaptive resonance and its connections with the internal mechanisms of cognitive activity are investigated. Considering the internal mechanisms of transformation of model representations of mental images, coactivation of connections in neural ensembles, concatenation of metastable states of model representations and adaptive resonance of neural activity, the authors come to the conclusion of close interdependence of similar dynamic processes, since their relationship can be caused by the same characteristics of internal cognitive mechanisms.

It can be expected that the understanding of the connection of adaptive resonance with the internal mechanisms of the cognitive space of memory will determine the understanding of the architecture of the cognitive process of thinking and the direction of further research, will provide additional arguments in favor of the awareness of the learning process as a process of developing the cognitive space of memory and activating the thinking of the individual. Such an understanding of the learning process will ensure further improvement of the technologies of professional and cultural development of the individual in Technical Universities.

The principles of adaptive resonance, on which the system architecture with self-organization properties was built, created the possibility of building and developing neural systems with self-regulation of a new type of artificial intelligence. [3,4]. There are well-known publications in which the authors use the principles of adaptive resonance theory when building models of biological neural networks, the dynamic properties of which, in their opinion, determine cognitive perception, cognition, and emotions, forming "adaptive intelligence." [5].

To date, numerous publications on the neurodynamic approach to the study of the activity of the neural system of the brain are known. [6]. This approach is based on the assumption that cognitive functions can be understood on the basis of the analysis of non-stationary processes and metastable states of the theory of dynamic systems and the hypothesis of functional modes of the cognitive space of memory.

Later, R. Makin and V. Lisin suggested that the stable state of the dynamic mode of neural activity should be considered as the "substrate equivalent of a psychological image." [7]. In the works of the authors [8,9], an attempt was made to study the internal mechanisms of the cognitive space. In 2019, a model of the mental space of the cognitive process of an individual was proposed [8].

Considering the cognitive cognitive process from the standpoint of the neurodynamic concept of the modes of localization of metastable structures in the phase space of the system, we present it as a process of creating a response to an external or internal stimulus. Creating a response to a stimulus is an intuitive processing of received information and is determined by the internal mechanisms of the neural system, so that the dynamics and direction of the process of creating a response are characterized as "cognitive unconscious" components. (For the first time, V. Allahverdov defined the internal mechanisms of the cognitive process as the "cognitive unconscious". [8]). The stimulus "turns on" the response creation mechanism, "setting the program for activating the fundamental modes of the cognitive memory space, restores the structures of neural ensembles, reproduces the localization regime of metastable structures and, accordingly, coded model representations of mental images that have signal-like elements" [8].

We define model representations of mental images as localized structures of metastable states in the phase space of eigenfunctions of a dynamic nonlinear problem in the neural environment. The phase space of activated localized structures is called the mental space of model representations of mental images. The mechanism of the creation of the response activates mental images of the cognitive memory space that are similar to the signal, restores the corresponding model representations, activating their mental space.

In the works [8,9] are discussed the internal mechanisms of the cognitive space of the cognitive process as the internal mechanisms of the mental space To such mechanisms we include mechanisms of coactivation of connections in the neural ensemble and concatenation of metastable structures of model representations of mental images. Along with the internal mechanisms of the cognitive memory space, the mechanisms of coactivation and concatenation, the adaptive resonance of the activated neural network makes a certain contribution to the process of formation and modification of metastable structures of model representations of the mental space.

The theory of adaptive resonance was developed for construct new types of neural network architecture models for the purpose of improving artificial intelligence systems. The proposed algorithms for the construction such architectures of neural networks model the rules of information perception by the neural system so that new information that entered into the system can be modified taking into account the existing context without the risk of destroying previously accumulated information.

Adaptive resonance of the neural network of the artificial intelligence system is defined as a dynamic process of system activation, during which the input signals of new information are matched with previously accumulated data. The coordination mechanism is due to the comparison and possible adjustment of the weight coefficients (connections) of the neurons involved in the process. Activation and subsequent adjustment of the weight coefficients is carried out when a signal is received at the input of the neural network, the image of which most closely matches the image "fixed to one of the neurons of this network." If, as a result of the agreement, the level of adjustment of the weighting coefficients reaches a certain threshold, the weighting coefficients of the activated neuron are modified, as a result of which the "information" attached to the neuron is supplemented with new data. The modification process is called adaptive resonance.

Adaptive resonance is the response of a dynamic system to a disturbance, as a result of which the system can acquire new properties. This understanding of adaptive resonance allows us to define it as a guiding component of the internal mechanisms of concatenation and coactivation of model representations in the cognitive process.

The research of cognitive processes of brain activity within the framework of the theory of nonlinear dynamics that has been applied to the neural environment has been extremely fruitful. In a nonlinear environment, the excitation of structures of different complexity, formed by the eigenfunctions of the dynamic problem, is determined by the parameters of the nonlinear environment. [5, 6]. The neural system of the brain is an environment in which a dynamic functional system is formed to perform cognitive functions. Sensory structures perceive and transform received signals, forming a set of active neurons. This neural environment is considered as a dynamic nonlinear system that forms dynamic regimes with metastable structures. A mathematical approach within the framework of dynamic systems to the research of certain brain functions with a high probability indicated that the implementation of brain functions is carried to the formation of regimes localization. As a result of homogeneous or heterogeneous transitions, the interaction of metastable structures can occur within the framework of one or close regimes localization.

Metastable regimes of a nonlinear dynamic problem are caused by solutions of a nonlinear eigenvalue problem that has a discrete spectrum. Eigenfunctions in the phase space of a dynamic problem correspond to metastable states that can create complex metastable structures under certain conditions. Metastable structures of a nonlinear dynamic problem in the neural environment form integrated model representations that carry information about the mental image of the signal. They are stored in the cognitive memory space in an encoded form, in the form of functional modes of a spatial signal or functional modes of a spatio-temporal signal. [10].

Studying the structure of the cognitive space of memory, we concluded that the fundamental units of this space are " Convolution" of functional modes of a certain category of mental images. "Convolution" can be considered as a system of functional modes, phase portraits of integrated model representations of the mental image of the signal. This system is structured so that each functional mode in this "Convolution" has many connections with functional modes of different levels. The existing system of connections of functional modes determines the most optimal structure of the "Convolution", which ensures minimal energy consumption in order to preserve the "Stability" of the cognitive memory space.

It is known that the cognitive space of memory, as the space of phase portraits of integrated model representations of mental images, is characterized by certain properties of "perception": plasticity, which is adaptation to new information, and stability, when new information should not destroy "the memory of old images".

Let's consider in more detail the mechanism of perception by the cognitive memory space of the coded mode representation of the mental image of the signal in the form of a functional mode. We will assume that the cognitive memory space is the space of functional modes. Perception by the cognitive memory space of a new functional mode is analogous to the disturbance of a system in a state of "stable equilibrium". In the system starts the process of searching for a new equilibrium state, which corresponds to checking the received information on "novelty". The result depends on the critical value of the modification level, which determines the probability of modification of the components of the disturbed system. If a mode is found for which the level of modification exceeds the critical value, then the search ends, the functional mode is modified under the "influence" of the functional mode of the new image. A similar process is defined as adaptive resonance.

If, as a result of the search, the level of modification is not higher than the critical value of modification, then the check on "novelty" of the functional mode of the new image is considered successful, the functional mode is perceived by the system as new and remains in the cognitive memory space.

The cognitive space of memory is the space of fundamental units. The fundamental unit is a "convolution" of functional modes, which unites the functional modes of a certain category of mental images. Therefore, the process of mapping the coded model representation of the mental image into the cognitive memory space, the space of fundamental units, is multi-step.

We will analyze the mechanism of perception of the coded model representation of the mental image by the cognitive memory space as a space of fundamental units. The process of displaying a model representation of a new image in the cognitive memory space starts the process of choosing a fundamental unit. We understand the fundamental unit as a "convolution" of functional modes, an "attractor" of functional modes. In such a case, the search of the choice can be completed by "capturing" the functional mode of any fundamental unit or by creating a new fundamental unit with the functional mode of a new image.

Capturing a new mode by a fundamental unit triggers a search mechanism on "novelty", comparison on "novelty" with functional modes of the selected "convolution". If the critical value of the level of modification with one of the "convolution" modes is overcome, then the selected mode is modified. In this case, we talk about adaptive resonance. If the critical value of the modification level is not overcome, then the check on "novelty" is considered successful, the functional mode of the new image remains in the "convolution" of functional modes.

Let's consider in more detail the process of checking on "novelty" of functional mods. Checking on "novelty" comes down to finding functional mode that has elements of similarity with the functional mode of the new image. The space of functional modes is the space of phase portraits of model representations of mental images. Therefore, checking on "novelty" determines the degree of similarity of model representations of mental images, the degree of similarity of mental images. The critical value of the level of modification determines the degree of similarity of functional modes, their mental images. For modification a mode is selected, for which the degree of similarity with the new mode is enough large.

The process of modification of the functional mode, previously embedded in the cognitive memory space, upon successful completion of the search for comparison with the phase portrait of the new image, is the process of modification of the integrated model representation. The dynamics of modification of the functional mode depends on the architecture of neural ensembles and the nature of neural connections that generate integrated model representations of the mental image. The activation of the functional mode corresponds to the restoration of neural ensembles in which the regimes of localization of metastable structures responsible for the formation of a mental image in the form of an integrated model representation are reproduced. A similar approach allows us to consider the dynamics of modification of a model representation of a mental image as a modification of neural ensembles, regimes of the localization , and, accordingly, metastable structures.

The internal cognitive mechanisms of the cognitive process are responsible for the formation of a modified model representation of a mental image, to which we attribute the coactivation of connections of neural ensembles and the concatenation of metastable states of model representations. In 2012, J. Fauconnier, a well-known cognitive linguist, put forward the hypothesis that in the cognitive space at the neural level, sets of neural structures are connected, by all indications, by the principle of coactivation. [15]. The mechanism of coactivation communication, which was first mentioned by J. Fauconnier, suggests that the joint presence in the environment of several certain components with the ability to interact significantly increases the activity of each component and determines the result of the interaction. mental adaptive resonance cognitive

The action of the mechanism of coactivation communication, which occurs in neural network systems with the activation of several neural ensembles, is reduced to the modification of the neural ensemble according to the "program" set by the neural structure of the new image. In the theory of spatio-temporal characteristics of the activity of neural network systems, the mechanism of concatenation of metastable neural network structures operates along with the mechanism of coactivation communication. E. Benderska and A. Pereshein proved the possibility of the appearance of new complex structures with the help of concatenation, separating some link of one structure and joining it to another structure. [8]. Modification of metastable structures can occur as a result of homogeneous or heterogeneous transitions, during the interaction of metastable structures of the same or close localization modes.

Thus, with the acquisition of new information (a new image), which is embedded in the cognitive space of memory in the form of functional modes, phase codes of integrated model representations of mental images, the cognitive space of memory is expanding, which determines the further cognitive development of the individual.

The study of the role of adaptive resonance in the expansion of the cognitive space of memory in the cognitive process in the paradigm of neurodynamic systems, taking into account the hypothesis of functional modes of the cognitive space of memory, showed that adaptive resonance can be considered as a guiding component of the mechanism of modification of functional modes. With adaptive resonance, the possibility of modifying the functional mode depends on overcoming the critical level of modification, which is defined by the authors as a measure of the similarity of functional modes. Is selected the mode for modification, for which the degree of similarity with the new one is quite large.

The process of modifying the functional mode is defined as the process of modifying the integrated model representation of the mental image. The modification of the model representation of the image previously embedded in the cognitive space of memory is defined as the modification of neural ensembles, their regimes of localization, and, accordingly, metastable structures. The dynamics of the reformation of the model representation of the image is determined by the characteristics of the internal cognitive mechanisms of the cognitive process.

Thus, a more detailed understanding of the connection of adaptive resonance with the internal mechanisms of the formation of the cognitive space will determine the direction of further research and understanding of the architecture of the cognitive thinking process, may provide additional arguments for understanding the learning process as a process of developing the cognitive space and activating the thinking of the individual. [16].

References

1. Grossberg, S. A. Path Toward Explainable AI and Autonomous Adaptive Intelligence: Deep Learning, Adaptive Resonance, and Models of Perception, Emotion, and Action Front Neurorobot. 2020 Jun 25;14:36.

2. Carpenter, G. A., Grossberg, S. A. Massively Parallel Architecture for a Self-organizing Neural Pattern Recognition Machine. Computer Vision, Graphics and Image Processing, 37(1), 54-115.

3. Carpenter, G. A., Grossberg, S. A. ART 2: Selforganization of Stable Category Recognition Codes for Analog Input Patterns, Applied Optics, Vol. 26, Issue 23, 1987, 4919-4930.

4. Carpenter, G. A., Grossberg, S. ART 3: Hierarchical Search Using Chemical Transmitters in Self-organizing Pattern Recognition Architectures, Neural Networks, Vol. 3, Issue 2, 1990,129-152.

5. Grossberg, S. A. The Resonant Brain: How Attentive Conscious Seeing Regulates Action Sequences that Interact with Attentive Cognitive Learning, Recognition, and Prediction Atten Percept Psychophys. 2019 Oct; 81(7):2237-2264.

6. Rabinovich, M. I., Muezzinoglu, M. K. Nonlinear Dynamics of the Brain: Emotion and Cognition, Phys. Usp., 53:4 (2010), 357-372.

7. Emelyanova, T.V. (2020) Robochii prostir piznavalnoho navchalnoho protsesu yak prostir mentalnykh konstruktiv kognityvnoho mekhanizmu myslenia [Worspace of The Cognitive Process as a Space of Mental Constructs of the Cognitive Mechanism of Thinking] Naukovi Zapysky. Seriia: Pedahohichni Nauky, Is.191, 74-77 [in Ukrainian].

8. Emelyanova, T.V. (2019) Mentalnii prostir yak kharakterystyka stupenia rozvytku kognityvnykh zdibnostei v navchalnomu protsesi [Mental Space as a Characteristic of the Degree of Development of Cognitive Abilities of the Individual in the Educational Process] Pedagogical Sciences: Theory, History, Innovative Technologies, Is.1 (85), 224-234 [in Ukrainian].

9. Emelyanova, T.V., Nesterenko, V. O. (2017) Pro mekhanizm aktyvizatsii piznavalnoho prostoru osobystosti v protsesi myslenia [About the Mechanism of Activation tf the Cognitive in the Thought Process] Pedagogical Sciences: Theory, History, Innovative Technologies, Is.3 (67), 165-175 [in Ukrainian].

10. Emelyanova, TV. (2020) Mekhanizm spryiniatia prostorovo-chasovoi informatsii v kohnityvnomu prostori osobystosti [About the Mechanism of Perception of the Spatial- temporal Information in the Cognitive Space of the Individual] Scientific Papers of Berdyansk State Pedagogical University. Series: Pedagogical sciences, Is.1, 258-267 [in Ukrainian].

11. Fauconnier, G. Mental Spaces. In D. Geeraerts, H. Cuyckens (Eds) The Oxford Handbook of Cognitive Linguistics.

12. Yarcho, T. (2016) Fundamentalizatsiia matematichnoi pidhotovky maibutnikh fakhivtsiv tekhnichnoho profilm u vyshchikh navchalnykh zakladakh [Fundamentalization of Mathematical Preparation of the Future Specialists of a Technical Profil at Higher Educational Establishments]. Monografiia. - Kharkov, 2016. - 284 p. [in Ukrainian].

Размещено на Allbest.ru