previous day
all days

View: session overviewtalk overview

10:00-11:30 Session Poster3: All posters will be on display in all poster sessions. Posters listed below should be presented by authors in this session only.
Extending the intelligence of the Pioneer 2AT mobile robot
PRESENTER: Michael Rudy

ABSTRACT. This paper describes the process of expanding the intellectual capabil- ities of a Pioneer 2AT skid-steer mobile robot by developing a module that allows using the ROS (Robotic Operating System) and its capabilities. This software and hardware system allow to quickly expand the capabilities of an outdated mobile robot, giving it access to modern algorithms e.g. SLAM. It also granted the sup- port for various attachments - lidars, stereo cameras, etc, autonomous navigation algorithms, data collection for training CNNs, as well as simulating the operation of these algorithms in the Gazebo. The module consists of two parts, the first is launched on the Arduino board, to which the wheelbase with the associated sen- sors is connected, the second part is the ROS node, which translates all the data coming from the Arduino into a format corresponding to the ROS interface.

Eligibility of English hypernymy resources for extracting knowledge from natural-language texts
PRESENTER: Oleg Sychev

ABSTRACT. A common subtask of knowledge acquisition from natural-language texts is classifying words and recognizing entities and actions in the text. It is used in the analysis of both scientific and narrative texts. Thesauri and lexical databases containing hypernymy relationship between synsets may be a useful resource for entity and action recognition. In this study, we compared the performance of three major English thesauri containing hypernymy relationship in different forms - WordNet, Roget's Thesaurus, and FrameNet - on 6 word-meaning categories that are used for the analysis of narrative and scientific natural-language texts. The results show that WordNet contains more words than FrameNet, and is more suitable for scientific texts, but FrameNet contains better-defined hypernyms and shows better precision for many narrative natural-language tasks, especially for verbs. Roget's Thesaurus performance is average between WordNet and FrameNet in most word-meaning categories Enhancing FrameNet by adding more lexical units to existing frames would allow creating a powerful resource for entity and action recognition in text analysis. Fixing WordNet problems require revising its system of hypernyms.

Specialized software tool for pattern recognition of biological objects
PRESENTER: Evgeny Levin

ABSTRACT. This paper describes the results of work on developing the mobile application for recognition of faces and other biological objects. The application is designed with a focus on loading external machine learning models over the Internet, which allows you to change the model without making any modifications to the application. With such realization, the application can be used in many cases. For example, at carrying out conferences: organizers just need to train a model and send out the link for its downloading to all participants of the conference without any changes in source code. Participants will be able to find out the in-formation about other members they are interested in, as well as contact them directly from the application, both by phone and by e-mail. Instructions are given for teaching your own face recognition model using the Microsoft Custom Vision cloud service, which allows you to train regardless of the power of your local computer. As an example, a classification model was trained and the fol-lowing assessments of recognition quality indicators were obtained precision: 97.8% and recall: 95.8%. In our future work we consider adding the functionality of emotion recognition based on the pattern recognition algorithm, described in this paper.

Accuracy of Expert Assessments in Evaluating Innovative Projects
PRESENTER: Pavel G. Gudkov

ABSTRACT. This article addresses the issue of improving the accuracy of expert assessments in evaluating innovative projects. The proposed approach to increase the accuracy of expert decisions is based on formalization of sub-criteria assessments using T. Saati's hierarchy analysis. Preliminary research shows the prospects of using the aforementioned approach. Further, the article discusses how its method can improve the accuracy of experts on the example of already existing methodological tools for evaluating innovation applications used by the Foundation for Assistance to Small Innovative Enterprises (FASIE). This work was supported by the MEPhI Academic Excellence Project (Contract No. 02.a03.21.0005).

Russian-language Neurosemantics: Clustering of Word Meaning and Sense from the Oral Narratives

ABSTRACT. This article is a part of a large-scale brain mapping project aimed at finding the correspondence of the semantic categories in oral Russian-language texts and the brain activity as measured in magnetic resonance scanner. The goal of present study in particular is to examine the nature of lexical semantic relations and find the appropriate lexical space, homeomorphic to the activation patterns in the brain. We present oral narratives, which described significant social problems from the first-person perspective, as stimuli and apply different annotation methods to encode the semantic information of words. Two approaches to annotation are described in the article: a dictionary method and a vector one. We also register fMRI signal and find clusters of words in input texts that have similar patterns of brain activation across subjects. These neurosemantic clusters are described in the article. Our results show that annotation by a list of features more strongly contributes to prediction of the observed activation patterns. These results also confirm the hypothesis of situational semantic representation in the brain.

Algorithm for constructing logical operations to identify patterns in data
PRESENTER: Larisa Lyutikova

ABSTRACT. Neural networks have proven themselves in solving problems when the input and output data are known, but the cause and effect relationship between them is not obvious. A well-trained neural network will find the right answer to a given request, but will not give any idea about the rules that form this data. The paper proposes an algorithm for constructing logical operations, in terms of multi-valued logic, to identify hidden patterns in poorly formalized areas of knowledge. As the basic elements are considered many functions of the multi-valued logic of generalized addition and multiplication. The com-bination of these functions makes it possible to detect relationships in the da-ta under study, as well as the ability to correct the results of neural networks. The proposed approach was considered for classification problems, in the case of multidimensional discrete features, where each feature can take k-different values and is equivalent in importance to class identification

Correlation of a Face Vibroimage Informative Parameters with Characteristics of a Person’s Functional State When Using VR and AR Technical Means

ABSTRACT. An analysis of the specifics of modern VR, AR, MR and XR systems use in educational, recreational and recovery processes is presented. The urgency of contin-uous monitoring of the current functional state (FS) and its psycho-emotional state (PES) using VR, AR, MR and XR technical means has been substantiated. Maintain-ing the state of health, as well as the required level of performance, are highlighted as the main requirements for the applied VR, AR, MR and XR means. The relevance of the use of infrared (IR) biometric technologies for the registration of the most in-formative human bioparameters, determining his current PES, has been substantiated. Analysis of IR vibraimage of a face is highlighted as the most promising technology for assessing its current PES. The possibility of determining the current FS on the basis of a set of bioparameters determining its PES is substantiated. For this purpose, the analysis of the correlation of the FS and the most informative bioparameters determined by the analysis of the IR vibraimage of the user's face by VR, AR, MR and XR means is given. To assess the FS in the study, it was proposed to use a value determined by the number of errors committed during periodic execution of specialized tests integrated into the VR, AR, MR, and XR scenario. Experimental data obtained during laboratory testing of the method confirmed the possibility of assessing FS based on a set of bioparameters measured during the processing and analysis of the IR vibraimage of the VR, AR, MR and XR user's face. The research results are of particular importance when training operators to control potentially hazardous facilities using VR, AR, MR or XR technologies, and, first of all, for the nuclear industry.

Algorithm for constructing logical neural networks based on logical various-valued functions

ABSTRACT. The intelligent control system, as a set of production rules, is implemented in the form of an various-valued logical function. The combined use of math-ematical logic and neural network methods gives the intelligent control sys-tem additional flexibility and the possibility of self-learning. In this paper we propose a method for representing various-valued logic function in a logical neural network. This logical neural network will keep the totality of cause-and-effect relationships identified using various-valued logic functions with-in a given specified area. These logic operations are implemented by special logic neural cells: conjunctors and disjunctors. The theorems given in this ar-ticle justify the possibility of constructing such neural networks. The method of proof of these theorems contains an algorithm for constructing logical neural networks for a finite number of steps.

Acoustic Pattern Recognition Technology Based on the Viola-Jones Approach for VR and AR Systems

ABSTRACT. The ability to solve problems of graphic images recognition in VR, AR, MR and XR systems is highlighted as one of the most important. The urgency of solving problems of recognition and classification of acoustic images has been substantiated, which will bring the quality of VR, AR, MR and XR systems closer to real reality (RR). Independent solution of graphic and acoustic patterns recognition problems using heterogeneous algorithmic and software tools is attributed to the disadvantages of modern systems. The study proposes an approach that allows the use of unified methodological and software tools for the simultaneous solution of graphic and acoustic patterns recognition problems. The proposed approach is based on converting acoustic information into graphic information using 2D-images of dynamic sonograms. This allows the recognition of acoustic patterns using unified algorithmic and software tools. It is proposed to use the Viola-Jones technology as such a unified tool. It is shown that the implementation of a two-stage determination of similarity measures of primitives and areas of the original image makes it possible to increase the speed of algorithms. For this purpose, at the first iteration, it is proposed to use not the graphic primitives themselves, but their coordinate projections. In the study, by analogy with Haar's features, parametrizable acoustic primitives were developed, presented in the classical graphical version, as well as in the form of coordinate projections

Review of fMRI methods in developmental stuttering and it's treatment

ABSTRACT. The article gives a summary overview of the most important studies of stuttering by using functional magnetic resonance imaging (fMRI) in the last 20 years. This review also highlights problems in the literature in terms of methodology and research areas. It presented an integrated approach and technique fMRI studies aimed at both the primary diagnosis of stuttering and stammering on the dynamics of the mechanisms in the treatment process.

Improvement of the technology of fMRI-experiments in the concealed information paradigm
PRESENTER: Kholodny Yuri

ABSTRACT. This work is a continuation of research on the creation of a forensic method for MRI diagnostics of hidden information in a person. The article presents some results of the first (technological) stage of these studies, during which the method of fMRI experiments was improved by using an MRI-compatible polygraph (MRIcP) within the framework of the information concealment paradigm. On the numerous experiments are shown: the efficiency MRIcP during fMRI; the usefulness of the methodological technique created through the use MRIcP that have improved the performance of an fMRI experiment; prospects of application of certain tests (in compliance with the strict requirements of forensic science) in the context of an fMRI study.

Expandable Digital Functional State Model of Operator for Intelligent Human Factor Reliability Management Systems
PRESENTER: Mikhail Alyushin

ABSTRACT. The relevance of the human factor (HF) reliability management is substantiated due to the reliable prediction of a possible change in the functional and psycho-emotional state (PES) of operators for controlling potentially hazardous objects (PHO). It is shown that PES modeling based on digital behavioral models (DBM) is a modern tool for such forecasting. Personalization of the DBM is carried out by taking into account individual bioparameters, which are registered, as a rule, using remote non-contact biometric technologies. The low reliability of the forecast in the event of emergency situations is highlighted as the main drawback of the existing DBMs. The main reason is the lack of reliability of the registration of bioparameters when using one or a limited number of biometric technologies. The relevance of the development of an expandable DBM, which allows to eliminate the indicated drawback, has been substantiated. The study proposes an expandable DBM that allows you to expand the range of processed biometric data. At the same time, it becomes possible to integrate biometric data obtained using remote non-contact technologies with data obtained using wearable biometric devices, such as, for example, bracelets. The developed DBM makes it possible, simultaneously with the implementation of the forecast, to monitor the health status of PHO personnel.

Comparative analysis of methods for calculating the interactions between the human brain regions based on resting-state fMRI data to build long-term cognitive architectures
PRESENTER: Alexey Poyda

ABSTRACT. Research in the field of constructing functional connectomes of the human brain at rest has been carried out over the past several decades. To build a functional connectome, it is necessary to evaluate the level of interaction between the regions that determine its functional elements. To determine this level, it is necessary to determine both the joint work of different brain areas, as well as establish of causal relationships between them. Currently, many different methods have been proposed for calculating the interaction of brain regions based on correlation, Granger causality, transfer entropy, coherence, mutual information, etc. Moreover, each of the methods depends on a number of parameters, which could affect the obtained result (for ex-ample, the size of the sliding window, frequency bands, model order, etc.). It is impossible to compare methods directly by the accuracy of evaluated connections, since we do not have a priori knowledge of functional connec-tions in the brain. Therefore, methods can be compared only by indirect measures. In this work, we compared many different methods according to the criterion of the stability of the results to small changes in the parameters of both the methods themselves (for example, the window size) and the in-put data (for example, the shift of the window in time or the shift of the brain region's boundaries in space). By stability we mean that small changes in the parameters will lead to small changes in the obtained estimates of the interaction. Currently, there is evidence in favor of the so-called dynamic connectivity of the brain regions, namely, microstates (from milliseconds to tens of sec-onds) that vary in time. However, since fMRI has a temporal resolution of about 2 seconds, here we focused on long-term architectures (400 seconds or more). This can also be explained by a number of reasons: 1) many meth-ods provide good estimates only with a sufficiently large time-series length (from 200 values or more); 2) we can expect the stability of long-term con-nections characteristics in space and in time, because the estimate is calcu-lated by averaging many functional interactions between brain areas that can be observed during the analyzed long-term period; 3) if the method does not show a stable result in the long-term analysis on averaged functional connections, this may indirectly indicate its unsuitability for the analysis of short-term conditions, while methods that showed good results on long-term analysis, might be further considered for constructing functional con-nectomes on a smaller time scale.

Collaborative creation and use of cognitive ontology-based domain information space for scientific research and learning
PRESENTER: Anton Anikin

ABSTRACT. The paper addresses the issues of knowledge management in learning and scientific research, in particular knowledge transfer from experts to consumers and the relevant processes. To solve these problems, we present a generalized methodology of intelligent support of decision making in knowledge management for scientific research and learning, supporting personalizing and collaborative creation and reuse of objects in domain information spaces. The domain cognitive information space is implemented as an ontological knowledge base using Semantic Linked Network as a formal basis for representing and visualizing the domain information. The paper describes methods and algorithms for representing, acquiring, and visualizing of knowledge based on the proposed model for creating and using domain information spaces. The proposed model and methods allow supporting decision making for knowledge management at the early stages of scientific research and in learning for creating a domain information space of the course, knowledge acquisition and skill formation, composing and adapting an educational trajectory, educational resources search, etc.

Neurophysiological features of neutral and threatening visual stimuli perception in patients with schizophrenia
PRESENTER: Sergey Kartashov

ABSTRACT. The authors of this work set a goal to study the features of visual perception of threatening stimulus associated with personal experience in patients with schizophrenia. As a target group were taken patients with a paranoid-hallucinatory syndrome. Healthy volunteers without mental disorders and neurological diseases were used as controls. During fMRI studies, threatening and neutral images were presented. The experiment was built on the principle of a block paradigm. As a result, statistical parametric maps were constructed for two groups of subjects and the results were compared among themselves. According to the obtained results, patients with schizophrenia show a decrease in the overall level of activation in all regions of the brain compared with healthy volunteers. This is most evident in Middle Temporal Gyrus (temporooccipital part Right), Inferior Temporal Gyrus (temporooccipital part Left and Right), Lateral Occipital Cortex (inferior division Left and Right), Temporal Occipital Fusiform Cortex Left and Right and Frontal Pole Left and Right.

Machine learning based on the principle of minimizing robust mean estimates

ABSTRACT. Abstract. The article considers the approach to the construction of robust methods and machine learning algorithms, which are based on the principle of minimizing estimates of average values that are insensitive to outliers. Proposed machine learning algorithms are based on the principle of iterative reweighting. Illustrative examples show the ability of the proposed approach and algorithms to overcome the effects of outliers.

Comparison between Coordinated Control and Interpretation Methods for Multi-Channel Control of a Mobile Robotic Device
PRESENTER: Timofei Voznenko

ABSTRACT. Control methods of the mobile robotic device can be divided into single-channel and multi-channel. A single-channel control method is a control method that uses a single data channel. Each control channel has its own operational features that can affect the quality of control. In order to take advantage of multiple control channels, multi-channel control is used. In multi-channel control information comes from different, heterogeneous, independent control channels. The task of the control system is to make a decision and choose valid (the most probable) command. There are two approaches to this problem: coordinated control and decomposition of multi-channel control into single-channel control. The decomposition method implies ignoring of command information from not selected channels and commands. The interpretation method allows using such command information to improve the quality of control of a mobile robotic device. In this paper, we consider coordinated control and interpretation methods and also compare implementations of these control methods for multi-channel control of a mobile robotic device.

Cognitive Architectures of Effective Speech-Language Communication and Prospective Challenges for Neurophysiological Speech Studies

ABSTRACT. The paper focuses on the importance of social cognitions and priors in natural cognitive architectures of an individual. The structures and content of perceptual-cognitive-metacognitive processes are analyzed using the material of natural speech-language communication related to early human ontogenesis. Metacognitive processes are defined as a property and an integral part of the cognitive system. The prospects of neurophysiological research are formulated, which are designed to clarify the distribution, the nature of connections and neuronal dynamics in the system of perceptual-cognitive-metacognitive processes that ensure effective speech communication at different periods of human development. The importance of the studies of the neural architectural solutions related to the processes of natural speech and language communication for the development of IT technologies which can fulfill the communicative needs and expectations of individual is emphasized.

Brain cognitive architectures mapping for neurosurgery: resting-state fMRI and intraoperative validation
PRESENTER: Maksim Sharaev

ABSTRACT. Despite the importance of experimental confirmation, the ability of wide range of brain mapping methods to discover brain cognitive architectures in most studies can’t be evaluated directly. Only in rare cases, when due to medical need, it is possible to conduct experiments during neurosurgical operations, is it possible to assess the accuracy of certain approaches. In this paper we evaluate, how well we can reveal brain cognitive architectures structure by established and novel approaches based on fMRI data, with special focus on resting-state fMRI, and how well these findings match cortical stimulation mapping data, which is a gold standard in neurosurgery. We illustrate our approach on three representative examples with different cognitive architectures mapping: brain motor network and language network, namely Broca and Wernicke areas. We found a significant correspondence between predicted maps and intraoperative data for both brain networks. This indicates that resting-state fMRI could be used as an additional source of information for neurosurgical planning, though its applicability to exploring and describing the whole variety of brain cognitive architectures for research purposes should be investigated in future.

Designing software for risk assessment using a Neural Network
PRESENTER: Anna Lebedeva

ABSTRACT. This article presents the results of the research in terms of using mathematical methods for the risk management process in the implementation of software de-velopment projects. Software development projects are not always implemented in a final form that meets the expectations of customers. Both internal and exter-nal factors can influence this. In this regard, the problems of risk management, which inevitably arise during the implementation of software development pro-jects, become particularly relevant due to the large uncertainty of the internal and especially external environment of enterprises. The introduction of a comprehen-sive approach to risk management allows the company to form an objective view of the current and planned activities of the organization, taking into account pos-sible negative events or new opportunities, anticipate risks and make decisions based on information about them, respond to risks in a timely manner and reduce the negative impact of risks in their implementation. Within the framework of this research, the risk assessment software is designed to assess the situation in the project, predict the future effectiveness of the project, and build scenarios to sup-port decision-making. Such software will allow to combine all the actions of the analyst in one tool, where all the information about the project and the external environment will be stored, updated and constantly used for training the Neural Network apparatus on which the software is designed. This work was supported by RFFI grant № 20-010-00708\20.

Analysis of using of neural networks for real-time process control
PRESENTER: Vasilii Volodin

ABSTRACT. Machine learning is one of the key technologies of the current scientific and technological revolution. Despite the fact that research in the field of "intelligent" control systems began in the last century, real-time control systems based on machine learning, specifically neural networks, began to be actively implementeed only in the past decade. In this paper, the authors analyze the current state of the problem of using real-time control systems based on neural networks.

Principles of design of a learning management system for development of economic skills for nuclear engineering education
PRESENTER: Gennady Baryshev

ABSTRACT. The dramatic changes in education that we face in 2020 and which are the result of the accelerating digital revolution indicate the need to develop new tools for education and training. Leading universities are developing their own Learning Management Systems (LMS) based on information and Internet technologies. However, in higher education there are certain educational tasks in the interdisciplinary field. One of them is the task of successfully developing atomic engineering education. In this article, we present the basic design principles and architecture features of a specific LMS for the development of economic skills for educa-tion in the field of nuclear technology.

Application of information measuring systems for development of engineering skills for cyber-physical edu-cation
PRESENTER: Gennady Baryshev

ABSTRACT. The new industrial revolution opening the way to the digital world requires further development of engineering education. Future engineers should obtain skills in the area of development of cyber physical (intellectual) systems. In National Research Nuclear University MEPhI we have examples of implementation of new engineering courses and programs for cyber physical education. In this paper we discuss the problems and results of application of information measuring systems which main purpose is for research and development needs, for educational tasks.

Design of a transcranial magnetic stimulation system with the implementation of nanostructured composites
PRESENTER: Gennady Baryshev

ABSTRACT. The method of transcranial magnetic stimulation has a number of confirmed ap-plications – not only for curing diseases, but also to develop neurocognitive abili-ties, such as language learning. Stationary treatment is rather expensive and not very convenient for many people. The trend so far is about development of mo-bile TMS systems. In this paper we present the results of our design of a tran-scranial magnetic stimulation system, that is different from other by application of nanostructured composites as a functional material. We discuss the features of such a design and options obtained by application of nanostructured materials.

11:30-13:00 Session Plenary4: Mostly Eastern Hemisphere
An Adaptive Network Model Covering Metacognition to Control Adaptation for Multiple Mental Models

ABSTRACT. Learning processes can be described by adaptive mental (or neural) network models. If metacognition is used to regulate learning, the adaptation of the mental network becomes it-self adaptive as well: second-order adaptation. In this paper, a second-order adaptive men-tal network model is introduced for metacognitive regulation of learning processes. The fo-cus is on the role of multiple internal mental models, in particular, the case of visualisation to support learning of numerical or symbolic skills. The second-order adaptive network model is illustrated by a case scenario for the role of visualisation to support learning mul-tiplication at the primary school.

An Adaptive Network Model for Pain and Pleasure through Spicy Food and its Desensitization

ABSTRACT. This paper aims to map out the adaptive causal pathways of processes underlying capsaicin consumption and the desensitization process of the TRPV1 receptor as a feedback loop together with pain and pleasure perception. In order to map out these causal capsaicin pathways, adaptive causal network modeling was applied, which is a way of modeling biological, neural, mental and social processes from an adaptive causal modeling perspective.

Cogmic Space for Narrative-Based World Representation

ABSTRACT. Representing a world or a physical/social environment in an agent’s cognitive system is essential for creating human-like artificial intelligences. This study takes a story-centered approach to this issue. In this context, a story refers to an internal representation involving a narrative structure, which is assumed to be a common form of organizing past, present, future, and fictional events and situations. In the artificial-intelligence field, a story or narrative is traditionally treated as a symbolic representation. However, a symbolic story representation is limited in its representational power to construct a rich world. For example, a symbolic story representation is unfit to handle the sensory/bodily dimension of a world. In search of a computational theory for narrative-based world representation, this study proposes the conceptual framework of a Cogmic Space for a comic-strip–like representation of a world. In the proposed framework, a story is positioned as middle-level representation, in which the conceptual and sensory/bodily dimensions of a world are unified. The events and their background situations that constitute a story are unified into a sequence of panels. Based on this structure, a representation (i.e., a story) and the represented environment are connected via an isomorphism of their temporal, spatial, and relational structures. Furthermore, the framework of a Cogmic Space is associated with the generative aspect of representations, which is conceptualized in terms of unconscious- and conscious-level processes/representations. Moreover, a proof-of-concept implementation is presented to provide a concrete account of the proposed framework.

Developing a Parallel Distributed Memory System of Stories: A Preliminary Report

ABSTRACT. The author's basic design concept of an artificial cognitive system is that story generation is an essential dimension of the mind. In this paper, a brief preliminary report on the development of a memory system is presented. The memory system is positioned as a partial system of a cognitive system. Its fundamental roles are to organize a world as a collection of stories and form the forces of generating stories. The memory system is designed as a parallel distributed system. In particular, each memory item—stories, events, concepts, and schemas—corresponds to an internal agent that operates automatically. These memory agents are dynamically organized in a network form through incorporating external simple narrative texts. This organizational process is performed via parallel distributed operations of connection formation and spreading activation among memory agents. The system is implemented in a proof-of-concept fashion. Several preliminary analyses of the system's behavior are also presented.

Application of Deep Reinforcement Learning to Decision-Making System based on Consciousnes

ABSTRACT. What is consciousness? Our research aims to achieve consciousness in computers and, in particular, focuses on a decision-making system based on consciousness. We have interested in a conscious decision-making system in an environment where multiple types of rewards and penalties exist. We know a method using a basis function and a method using a penalty avoidance list for this problem. Though the method using the penalty avoidance list is considered promising compared to the former, it has the problem that when all actions are registered in the avoidance list, the actions are selected at random. In this paper, we propose a method for selecting actions using deep reinforcement learning in order to avoid such random selection as much as possible. The effectiveness of the proposed method is confirmed by numerical experiments.

Does change in ethical education influence core moral values? Towards culture-aware morality model
PRESENTER: Jagna Nieuwazny

ABSTRACT. In this study, we focus on ethical education as a means to improve artificial companion’s conceptualization of moral decision-making process in human users. In particular, we focus on automatically determining whether changes in ethical education influenced core moral values in humans throughout the century. We analyze ethics as taught in Japan before WWII and today to verify how much the pre-WWII moral attitudes have in common with those of contemporary Japanese, to what degree what is taught as ethics in school overlaps with the general population’s understanding of ethics, as well as to verify whether a major reform of the guidelines for teaching the school subject of “ethics” at school after 1946 has changed the way common people approach core moral questions (such as those concerning the sacredness of human life). We selected textbooks used in teaching ethics at school from between 1935 and 1937, and those used in junior high schools today (2019) and analyzed what emotional and moral associations such contents generated. The analysis was performed with an automatic moral and emotional reasoning agent and based on the largest available text corpus in Japanese as well as on the resources of a Japanese digital library. As a result, we found out that, despite changes in stereotypical view on Japan’s moral sentiments, especially due to historical events, past and contemporary Japanese share a similar moral evaluation of certain basic moral concepts, although there is a large discrepancy between how they perceive some actions to be beneficial to the society as a whole while at the same time being inconclusive when it comes to assessing the same action’s outcome on the individual performing them and in terms of emotional consequences. Some ethical categories, assessed positively before the war while being associated with a nationalistic trend in education have also disappeared from the scope of interest of post- war society. The findings of this study support suggestions proposed by others that the development of personal AI systems requires supplementation with moral reasoning.

14:30-16:30 Session Plenary5: For All Participants
A model of top-down attentional control for visual search based on neurosciences
PRESENTER: Natividad Vargas

ABSTRACT. Visual attention is an essential and critical mechanism that allows humans to select the most relevant visual information of potential interest to focus on certain aspects of the environment. There are several proposals to model visual attention. However, those models only describe behaviors in simple tasks like free visualization. In more complex tasks such as visual search, it requires attention processes to guide behavior. Attentional control provides these mechanisms. Through which, top-down (goal-directed) information is represented and configured according to the various constraints and dynamics of task processing. In this article, we describe a task-dependent approach to model attentional top-down control based on neuroscience. We present a general conceptual model of visual attention. We describe its three main components and their relationship with other cognitive functions. Also, we show more detailed information about the flow of information that our model follows using a simple guided search case study. Our proposal intends to be the basis to treat top-down attentional information in a broader cognitive architecture. We find that the existence of IPS templates provide a general and biologically inspired representation for relevant objects. Our results show that the proposed model is significantly more consistent and explanatory in information processing compared to other state-of-the-art models.

A proposal for an auditory sensation cognitive architecture and its integration with the motor-system cognitive function.

ABSTRACT. The auditory system is capable of producing a wide range of information through the acquisition and perception of the vibrations present in the environment, even when the receptor is not directly facing the stimulus's source. Said information can be crucial for survival and useful for a variety of systems like the visual system and the motor system. Despite that, the quantity of studies involving the auditory system or its interactions with other systems is limited, even though anatomical evidence recognizes this relationships' existence. In this work, we study its interaction with the motor system. A bio-inspired model that explores the relationship between the auditory and motor systems, grounded on neuroscientific research, is presented to address this proposal. To validate our proposal, a case study in which we endow a virtual entity with our proposed model. Then, we ask both a group of persons and the virtual creature to compute and face towards the direction were the sound was originated.

Toward ethical cognitive architectures for the development of artificial moral agents

ABSTRACT. New technologies based on artificial agents promise to change the next generation of autonomous systems and therefore our interaction with them. Systems based on artificial agents such as self-driving cars and social robots are examples of this technology that is seeking to improve the quality of people's life. Cognitive architectures aim to create some of the most challenging artificial agents commonly known as bio-inspired cognitive agents. This type of artificial agent seeks to embody human-like intelligence in order to operate and solve problems in the real world as humans do. Moreover, some cognitive architectures such as Soar, LIDA, ACT-R, and iCub try to be fundamental architectures for the Artificial General Intelligence model of human cognition. Therefore, researchers in the machine ethics field face ethical questions related to what mechanisms an artificial agent must have for making moral decisions in order to ensure that their actions are always ethically right. This paper aims to identify some challenges that researchers need to solve in order to create ethical cognitive architectures. These cognitive architectures are characterized by the capacity to endow artificial agents with appropriate mechanisms to exhibit explicit ethical behavior. Additionally, we offer some reasons to develop ethical cognitive architectures. We hope that this study can be useful to guide future research on ethical cognitive architectures.

Bioinspired model of short-term satiety of hunger influenced by food properties in virtual creatures

ABSTRACT. The behavior of the human is continually changed as a consequence of various drives which human is predisposed also of your survival instinct. Among the basic drives of the human, there are the physiological needs and is precisely the hunger that motivates the food intake to get the energy that the body requires via food. The regulation of hunger allows to stop the food intake by means of the homeostatic and hedonic control which are influenced by the food properties. The process that consists of ending the food intake is known as short-term satiety and is important because it limits the amount of food intake; Otherwise, an over-consumed affect the organism functioning negatively. In this paper, we propose a conceptual model for the generation of short-term satiety behaviors based on neuroscientific evidence for virtual creatures. The conceptual model proposed is implemented in a distributed system—a virtual creature endowed with this implementation is placed in a virtual environment to analyze its behavior. The analysis shows how the virtual creature modifies its hunger level (behavior) based on food's properties. The results show the execution of the process when the creature interacts with the environment.

An expanded model for perceptual visual single object recognition system using expectation priming following neuroscientific evidence
PRESENTER: Ivan Axel Dounce

ABSTRACT. Under numerous circumstances, humans recognize visual objects in their environment with remarkable response times and accuracy. Existing artificial visual object recognition systems are still to surpass human vision, specially in its universality of application. We argue that modeling the recognition process in an exclusively feedforward manner hinders those systems' performance. To bridge that performance gap between them and human vision, we present a brief review of neuroscientific data which suggests that recognition can be improved by considering an agent's internal influences (from cognitive systems that peripherally interact with visual-perceptual processes). Then, we propose a model for visual object recognition which uses these systems' information, such as affection, for generating expectation to prime the object recognition system, thus reducing its execution times. Later, an implementation of the model is described. Finally, we present and discuss an experiment and its results.

Decision-making bioinspired model for target definition and “satisfactor” selection for physiological needs

ABSTRACT. Every person, from an early age, has to make decisions to resolve situations that arise in life. In general, different people make different decisions in the same situation, since decision-making takes into account different factors such as age, emotional state, experience, etcetera. We can make decisions about situations that we classify as: more important than others, routine, unexpected, or trivial. However, making the correct decision(s) in a timely manner for these situations is one of the most complex and delicate challenges that human beings face. This is due to the arduous mental process required to be carried out. Providing such behavior to a virtual entity is possible through the use of cognitive architectures or CA. CAs are the approach used to model human intelligence and behavior. This paper presents a useful bioinspired decision-making computational model to satisfy the physiological needs of hunger and thirst. Our proposal considers as a black box other cognitive functions that are part of a general CA (we name Cua ̄yo ̄llo ̄tl or brain in Nahuatl). In the proposed case study, it is proved that the decision-making process plays an essential role in determining the objective and selecting the object that satisfies the established need.

Declarative Working Memory: A Bio-Inspired Cognitive Architecture Proposal
PRESENTER: Luis Martin

ABSTRACT. Memory is considered one of the most important functions since it allows us to code, store and retrieve knowledge. These qualities make it an indispensable function for a virtual creature. In general, memory can be classified based on the durability of the stored data in working memory and long-term memory.

Working memory refers to the capacity to maintain temporarily a limited amount of information in mind, which can then be used to support various abilities, including learning, reasoning, planning and decision-making. Unlike short-term memory, working memory is not only a storage site, but it is also a framework of interacting processes that involve the temporary storage and manipulation of information in the service of performing complex cognitive activities. Declarative memory is a type of long-term memory related with the storage of facts and events.

This research focuses on the development of a cognitive architecture for the type of working memory that maintains and manipulates declarative information. The construction of the model was grounded in theoretical evidence taken from cognitive sciences such as neuroscience and psychology, which gave us the components and their processes.

The model was evaluated through a case study that covers the encoding, storing, and retrieval stages. Our hypothesis is that a virtual creature endowed with our working memory model will provide faster access to the information needed for the ongoing task. Therefore, it improves the planning and decision-making processes.

A proposal of bioinspired motor-system cognitive architecture focused on feedforward-control movements

ABSTRACT. The objective of this article is to present a conceptual motor-system cognitive architecture inspired in the human nervous system, anda cognitive architecture focused on the voluntary movement controlledby feed-forward. The article first focus on describing the brain cortexareas that compound the motor system, presenting the supplementarymotor area (SMA), premotor cortex and primary cortex, the tasks thatthese cortices do, and how them works together. Then, it is presenteda cognitive architecture based on the information presented. In this sec-tion, it is described the areas in a computational level (functions, andalgorithms used). Finally, it is presented a study case where the cognitivearchitecture proposed is used to execute a voluntary-movement-by-feed-forward-control task.

17:00-19:30 Session Plenary6: For All Participants. This session will also include the ICOM Workshop. Session will end with a joint discussion panel.
Applying Independent Core Observer Model Cognitive Architecture to a Collective Intelligence System

ABSTRACT. This paper shows how the Independent Core Observer Model (ICOM) Cognitive Architecture for Artificial General Intelligence (AGI) can be applied to building a collective intelligence system called a mediated Artificial Superintelligence (mASI). The details include breaking down the ICOM implementation in the form of the mASI system and the general performance of initial studies with the mASI. Details of the primary difference between the Independent Core Observer Model Cognitive Architecture and the mASI architecture variant include inserting humanity in the contextual engine components of ICOM, creating a type of collective intelligence. Humans can ‘mediate’ new system-generated thinking keeping the thought process accessible and slow enough for humans to oversee and understand. This also allows the modification of emotional valences of the thought process of the mASI system to help the system generate complex contextual models (knowledge graphs) of new ideas and which speeds up the learning process. With the humans acting as control rods in a reactor and emotional drivers, the mASI system maintains safety where the system would cease to function if humans walked away.

Causal Cognitive Architecture 1: Integration of Connectionist Elements into a Navigation-Based Framework

ABSTRACT. The brain-inspired Causal Cognitive Architecture 1 (CCA1) tightly integrates the sensory processing capabilities found in neural networks with many of the causal abilities found in human cognition. Sensory input vectors are processed by robust association circuitry and then propagated to a navigational temporary map. Instinctive and learned objects and procedures are applied to the same temporary map, with a resultant navigational signal obtained. Navigation can similarly be for the physical world as well as for a landscape of higher cognitive concepts. Causality emerges from the architecture, with good explainability for causal decisions. A simulation of the CCA1 controlling a search and rescue robot is presented with the goal of finding and rescuing a lost hiker within a grid world.

E-governance Experimental Framework Using a Mediated Artificial Superintelligence (mASI) System Research Study


Proposed Collective ICOM-based post-scarcity/post-capital networked communities

ABSTRACT. Among the myriad anthropogenic trends threatening our biosphere, several—including climate change, pollution, species extinction, and disease—are amenable to direct science- and technology-based intervention. However, non-environmental trends that impact the overwhelming majority of humans—specifically, extensive poverty, migration and homelessness brought about by wealth and wage disparity, resource scarcity, population growth, and unemployment—require technology- and collaboration-based interdisciplinary and transdisciplinary solutions. These latter issues can be attributed to current socioeconomic practices, which in turn are behavioral expressions of H. sapiens evolutionary neurobiology as a social species with an alpha-dominant hierarchy and a strong in-group/out-group bias—an aspect of human neurobiology that often results in mechanisms such as philosophy, political rhetoric, regulation, and legislation rendering alternative progressive political economies temporary or ineffectual. In addition, most alternative political economies share the assumption that capital is necessary to societal structure and function, and so have been limited in their progressive potential or replaced by capitalism. At the same time, more future-forward post-scarcity/post-capital systems remain theoretical. To that end, Transinopia (meaning beyond scarcity in one translation from the Latin trans inopia) differs in being a post-scarcity/post-capital economic system explicitly intended to be instantiated in a multiyear proof-of-concept field trial. Specifically, Transinopia encompasses the design, construction, and inhabitation of a network of technology-augmented, cooperation-based, self-sufficient enclaves embedded within a capitalism-based polity. Structured as a controlled scientific experiment, Transinopia will demonstrate the degree to these communities are structurally and functionally viable as well as how robustly their inhabitants (selected using a clinical trial model) thrive without wage-based labor or governmental monetary support.

Moreover, by incorporating an Independent Core Observer Model (ICOM) . , , , , , , —a cognitive architecture designed to produce complex internal subjective emotional experience to drive motivation, goals and all decisions—Transinopia will serve as a unique intuitive human-analogous Collective Artificial General Intelligence (cAGI), which is designed to provide continuous multiple human interactions, a controlled testbed for novel technologies, and a shared collaborative worldview. Moreover, if the Transinopia system proves viable, it will establish an environment that encourages critical thinking, empathic, and prosocial behavior, while the resultant empirical data will support the creation of larger-scale enclave networks.

Methodologies and Milestones for The Development of an Ethical Seed
PRESENTER: Kyrtin Atreides

ABSTRACT. With the goal of reducing more sources of existential risk than are generated through advancing technologies, it is important to keep their ethical standards and causal implications in mind. With sapient and sentient machine intelligences this becomes important in proportion to growth, which is potentially exponential. To this end, we discuss several methods for generating ethical seeds in human-analogous machine intelligence. We also discuss preliminary results from the application of one of these methods in particular with regards to AGI Inc’s Mediated Artificial Superintelligence named Uplift. Examples are also given of Uplift’s responses during this process.

The Enactive Computational Basis of Cognition and the Explanatory Cognitive Basis for Computing

ABSTRACT. The computational theory of cognition, also known as computationalism, holds that cognition is a form of computation. Thus, computation can be seen as a or the explanatory basis for cognition. The goal of this paper is to address two key aspects here: A) Computing systems are traditionally seen as representational systems, but functional and enactive approaches are supporting non-representational approaches to computing; B) Recently, a sociocultural theory against computationalism has been proposed with the aim of reducing computing to cognition, but we consider that cognition and computation are based on action, that cognition is just a form of computing and that cognition is the explanatory basis for computation. In this article, we state that: 1. Representational theories of computing recurring to intentional content run into metaphysical problems, which are unsustainable to characterize computing and cognition in nature. 2. Functional non-representational theories of computing and cognition do not incur this metaphysical problem when describing computing in terms of the abstract machine. 3. Functional theories are consistent with enactive theories of computing in describing computing machines not in a strictly functional way, but especially in terms of their organization. 4. This paper also claims that the theory of enactive or autopoietic cognition is consistent with the computational basis of cognition in describing Turing machines as functionally and organizationally closed systems. 5. It is here defended, the computational basis of cognition and the cognitive explanatory basis for computing, arguing that computer science is developed in the human linguistic domain, then as a product of human socionatural normative practices. The paper concludes by supporting that computational models are about actions and that computing is in action. Furthermore, cognition is just one form of computing in the world and the explanatory basis for computation.

Approaching the Psychology of AI


JVRT Closing Discussion

ABSTRACT. This discussion happened at the end of the ICOM workshop, which was the last session of the BICA*AI 2020 Joint Virtual Reality Track. Included authors are the main participants of the discussion. The abstract was submitted in order to make the item available for the automatically generated program.

20:00-21:00 Session Poster4: All posters will be on display in all poster sessions. Posters listed below should be presented by authors in this session only.
Selection of a Friction Model to Take into Account the Impact on the Dynamics and Positioning Accuracy of Drive Systems
PRESENTER: Sergey Misyurin

ABSTRACT. The problem of choosing a friction model for solving the problems of controlling positional systems, primarily with a pneumatic drive, is discussed. Due to their high dynamics, good towing capacity and relatively low price, pneumatic positioning systems are an attractive alternative to electric drives. However, the use of pneumatic systems involves some difficulties caused by the nonlinearities of its individual elements, in particular the flow characteristics of the servo valve, the compressibility of the working fluid, and also the friction acting on the piston. The main goal of this work is to analyze the stability in the interaction of the energy and control units under the influence of friction forces represented by various models. The Karnopp model was considered as one of the models, which has the advantage in describing the interaction with the friction forces in the transition from the state of rest to motion and vice versa.

Applicative model to bring-in conceptual envelope for computational thinking with information processes

ABSTRACT. In modern research, computing is understood as the study of natural and artificial information processes, which, in turn, are recognized as ubiquitous. Information processes are manifested through a variety of external forms, based on computational communication. Communication consists in the fact that the parties involved exchange process messages, and the same process message can be expressed in different sentences and vice versa, the same sentence can express different process messages. The ability to identify the same process in different sentences is based on the hypothesis of the stratification of each communication language into two languages: - a genotypic language for unambiguous recording of the content of processes; a phenotypic language with a variety of means for expressing the same process. The implementation of the connection of both languages by embedding in an applicative computational model equipped with a convertibility relation is proposed. The defended arguments boil down to the following: (1) the computation cannot be eliminated not only from the research method, but also from the subject of the study - what is being studied and (2) the real value of computer science lies in those conclusions that experts can draw based on their expertise using a fairly rich and deep system of reasoning. For example, the most fruitful Karoubi envelope features for computing are discussed.

Lateralization in Neurosemantics: Are Some Lexical Clusters More Equal Than Others?
PRESENTER: Zakhar Nosovets

ABSTRACT. In the study, we have implemented neurosemantic analysis to find brain’s voxel-wise representations of words in Russian spoken narratives and their asymmetries in the brain. 25 subjects were listening to five stories, first person narratives of dramatic events, while their brain activation was registered by 3T functional magnetic resonance imaging (fMRI). Seven best subjects in terms of their engagement and objective control of brain reaction were selected for further analysis. 12 lexical clusters were found, with different semantics – from time-and-space concepts to human actions and mental states. Cluster “experience” was the only one that demonstrated a slight right-sided lateralization. For other clusters, brain localization was either symmetrical or had a clear left-sided bias. Our results support the view of non-modular and widely distributed nature of semantic representations, not limited to the activity of structures in the temporal and frontal lobes. These results also demonstrate that the right hemisphere can be massively involved in representation of “reflexive” part of inner lexicon.

Friction model identification for dynamic modeling of pneumatic cylinder
PRESENTER: Vladimir Ivlev

ABSTRACT. Friction is one of the main nonlinear properties that makes pneumatic actua-tors difficult to control and reduces their energy efficiency. Many phenome-nological friction models are used to describe pneumatic cylinders dynamic behavior, including in pre-sliding zone. These models maintain many un-known empirical parameters (maybe 7 or more). Expensive test equipment and special mathematical methods are required to define the parameters of friction model. But the results may vary significantly for one size cylinders of various manufactures. This paper presents the results of determination the Stribeck friction model parameters based on limited experimental data and procedure of vector identification which implemented in the software com-plex MOVI (Multicriteria Optimization and Vector Identification). Results were obtained for two types of piston seals materials: NBR and PTFE com-posite. The minimal value of the piston stable speed for single action cylin-ders with these seals was estimated.

Remote laboratory-based classes at the department of general physics of NRNU MEPhI under the condition of the COVID-19 pandemic.

ABSTRACT. It is being discussed the organization of the laboratory-based classes and presentation by students their results at the department of general physics of NRNU MEPhI under the condition of the lockdown caused by the COVID-19 pandemic.

Applying a logical derivative to identify hidden patterns in the data structure

ABSTRACT. The paper proposes a method for assessing the significance of individual characteristics of recognized objects. The totality of objects and their charac-teristics is represented by the structure and weight coefficients of a trained -neuron. The specified -neuron correctly processes objects of the sub-ject area, which may not be explicitly represented. It is known that when us-ing the neural network approach, the logical rules for decision making by the neural network remain hidden to the user. The proposed method for con-structing the decisive function allows us to identify these logical rules of a correctly functioning -neuron. To assess the significance of the character-istics of objects, a logical derivative is used. Which shows how the decisive function will change its value if one or more characteristics of the objects change. That will allow us to conclude about the most important properties of the subject area under consideration. This is especially important when data is incomplete, fuzzy, or distorted due to information noise.

Synthesis of parameterized families of correctly functioning sigma-pi neurons

ABSTRACT. Abstract. A constructive method and an algorithm for constructing a parametrized family of $\Sigma\Pi$-neurons that correctly function on a given training set and have, in a certain sense, a minimal structure are proposed.

Logical circuits of a ΣΠ-neuron model

ABSTRACT. This article presents two logical circuits of a ΣΠ-neuron model. Each circuit is divided into two layers, realizing a multiplicative and additive operation. The circuits are built on the basis of digital and analog elements. The layer that performs the multiplicative function is identical in both circuits. The ad-ditive layer in the digital circuit is realized due to the digital adder. In the hy-brid circuit, it is realized by means of DACs and an analog voltage adder. To implement the threshold function in the hybrid circuit used a comparator. In the digital circuit used a digital comparator. The use of analog elements for the implementation of the summation functions has the advantages of speed over a digital adder, however, it loses in accuracy. The results obtained can be used to construct ΣΠ-neural networks which can be included in the hard-ware and software of mechatronic devices.

Mathematical Methods for Solving Cognitive Problems in Medical Diagnosis
PRESENTER: Tatiana Semenova

ABSTRACT. Complex system exploring at initial stage of includes fixation of basic data blocks, problem categories, and other. At this stage, an adequate description uses gestalt-like patterns for the future system, its components and their functioning. In the structure of such system, functional and logical relationships can be found between the components or processes. Such a complex system is a living organism (for example, a patient). A skilled patient-treating physician can discover functional and logical associations between the components or processes in the studied system. In some difficult cases, doctors cannot explain their decisions and actions. They might use it practically, but cannot formulate verbally. For these cases, mathematician I.M. Gelfand has proposed a method of diagnostic games. The game is a cognitive research to reveal the doctor's intuitive action plan for specific case of the patient's treatment. Working together, the mathematician and physician can formulate a verbal case description. The objective of this study is to extract strict formalized elements from the doctor's gestalt perception: the rules of diagnostic decision. In order to analyze the intuitive actions of the specialists, the authors propose a mathematical language and technologies based on non-numerical statistics and three-valued logic. The language helps us detect and solve such cognitive problems. The collaboration with doctors allows us to create clear diagnostic rules based on the latent knowledge of an experienced specialist. The article provides a brief description of the method used for solving the problems of practical medicine.

Data-Driven Model for Emotion Detection in Russian Texts
PRESENTER: Alexander Naumov

ABSTRACT. An important task in the field of automatic data analysis is detecting emotions in texts. The paper presents the approach of emotional analysis of text data in Russian. To conduct an emotional analysis, a method was created based on vector representations of words obtained by the ELMo language model, and subsequent processing by an ensemble classifier. To configure and test the created method, a specially prepared dataset of texts for five basic emotions -- joy, sadness, anger, fear, and surprise -- is used. The dataset was prepared using a crowdsourcing platform and a home-grown procedure for collecting and controlling annotators' markup. The overall accuracy is 0.78 (by the F1-macro score), which is currently the new state of the art for Russian. The results can be used for a wide range of tasks, for example: monitoring social moods, generating control signals for mobile robotic systems, etc.

Ensembling SNNs with STDP learning on base of rate stabilization for image classification
PRESENTER: Alexey Serenko

ABSTRACT. The possibility of solving real-vector classification tasks based on a simple spiking network with Spike-Timing-Dependent-Plasticity (STDP) learning was shown in our previous works. In this paper, that method is extended by aggregating neurons into ensembles, and validated on an image recognition dataset. The method is based on a one-layer network of neurons with STDP-plastic inputs receiving pixels of input images encoded with spiking rates. This work considers two approaches for aggregating neurons' output activities within an ensemble: by averaging their output spiking rates (i. e. averaging outputs before decoding spiking rates into class labels) and by voting with decoded class labels. Ensembles aggregated by output frequencies are shown to achieve a significant accuracy increase up to 95\% (by F1-score) for the Optdigits handwritten digit dataset, and is comparable with conventional machine learning approaches.

Hierarchical Deep Q-Network from Imperfect Demonstrations in Minecraft
PRESENTER: Alexey Skrynnik

ABSTRACT. We present Hierarchical Deep Q-Network (HDQfD) that won first place in the MineRL competition. The HDQfD works on imperfect demonstrations and utilizes the hierarchical structure of expert trajectories. We introduce the procedure of extracting an effective sequence of meta-actions and subgoals from the demonstration data. We present a structured task-dependent replay buffer and an adaptive prioritizing technique that allow the HDQfD agent to gradually erase poor-quality expert data from the buffer. In this paper, we present the details of the HDQfD algorithm and give the experimental results in the Minecraft domain.

Self and Other Modelling in CooperativeResource Gathering with Multi-AgentReinforcement Learning
PRESENTER: Vasilii Davydov

ABSTRACT. In this work, we explore the application of the Self-other-Modelling algorithm (SOM) to several agent architectures for the collab-orative grid-based environment. Asynchronous Advantage Actor-Critic(A3C) algorithm was compared with the OpenAI Hide-and-seek (HNS)agent. We expand their implementation by adding the SOM algorithm. As an extension of the original environment, we add a stochastic initialization version of the environment. To address the lack of performance in such an environment by all versions of agents, we made further improvements over the A3C and HNS agents, adding the module dedicated to the SOM algorithm. This agent was able to efficiently solve a stochastically initialized version of the environment, showing the potential benefits ofsuch an approach.