ABSTRACT. The work is devoted to GPU-based high performance realization of algorithm for wavelet phase synchronization. Wavelet phase coherence was applied for analyzing brain activity in states with different degrees of mental and sensory attention. In the analysis of electroencephalographic correlates of mental states, as a rule the focus is on the analysis of the spectral power of a quasi-stationary EEG or task-related power of time-frequency EEG spectra. The analysis of the wavelet phase coherence provides additional information on the organization of brain activity, but time consuming in estimation. Fast implementation can simplify the use of this method in practice.

ABSTRACT. This article examines a minimalist mechanism in a simple language game to acquire the use of two-word sentences consisting of a first- or second-person pronoun as the subject and a verb as the predicate. In an experiment, a learner agent with minimalist architecture learned to select the subject and verb while interacting with a caretaker agent. The assumptions and implications of the experiment are discussed.

ABSTRACT. What is human self? Some argue that there is no such thing as self. However, the subjective feeling that “I am writing these words” makes it hard to deny the existence of the self. We assume that as long as there is the term “self,” there must be some collection of neural networks that represents the concept of the term. Although the whole picture is still a mystery, we have taken a step forward to unraveling the mystery by introducing the idea that “the emergence of a new behavior that prioritizes the body underlies the rise of the self.” When performing imitation behavior, a person can encounter a situation in which he feels pain and tries to avoid it. In this instance, the person engages in two types of behavior almost simultaneously, which are in conflict with each other. Also in this instance, it is assumed that the person gives priority to the safety of his own body and reflexively chooses to respond with avoidance behavior. However, as the imitation behavior continues, the process of imitation and avoidance is repeated many times, making it increasingly difficult to ensure the safety of the body. To address this scenario, we have come up with an idea that enables a conscious system to generate a new rational behavior—that is, voluntarily stop the imitation behavior. We consider that the generation of this new behavior is a significant process that can explain the first step for the development of the self.

ABSTRACT. We have successfully simulated the cognitive process in looking at Rubin’s Vase. Rubin’s Vase is an ambiguous image developed by the Danish psychologist Edgar Rubin. The ambiguous image allows the viewer to interpret it in more than one way. The Rubin’s Vase image used in this study depicts a vase in the center in a way that its contour matches the human profile, allowing the viewer to interpret the image as either “a vase in the center” or “two faces looking at each other.” The program we have developed is designed to enable the internal representation systems to change their responses from moment to moment according to changes in input data and to internal knowledge.

ABSTRACT. Stalking behavior, characterized by specific and dreadful acts such as persistent following, repeated sending of unwanted gifts, lack of sympathy, and even willingness to kill the victim, has become a serious concern in modern society. However, since stalking behavior has something partly in common with any criminal behavior, further discussion of this topic is considered helpful in determining factors behind various criminal behavior. Thus, there is an urgent need to investigate how this mysterious and dangerous behavior arises. There are three main types of stalking behavior: “rejected,” “resentful,” and “other (intimacy seeker and incompetent suitor).” First, we focused on a conscious model of the rejected type, as it is considered the most typical. The artificial conscious model is built to represent a process in which a conflict of concepts arising in the Reason Subsystem is progressively reconciled by the Association Subsystem.

ABSTRACT. We evaluate human grouping of everyday objects for a psychologically inspired knowledge representation based on prototype theory. Our overall aim is to develop a knowledge representation system that one day could be used by kitchen robots. We conducted a study in which participants had to sort different kitchen objects into a digital kitchen. We chose a kitchen as a use case, since people have to tidy up dishes every day. We calculated an overall similarity matrix and identified object groups whenever at least half of the participants put two items on the same shelf. Out of these object categories we calculated the respective prototype. We then tested the similarities of all categories to all prototypes, which turned out to be reasonable.

ABSTRACT. In this paper, we discuss the possibilities of integration of the Robot Dream paradigm with Anthropomorphic Social Agent (ASA). Anthropomorphic Social Agent is essentially a user interface specifically designed for simulating the realistic human-to-human interaction. The main idea behind the Robot Dream paradigm is to provide highly realistic social behavior when resources of the machine are limited by outsourcing resource-intensive computation to the cloud. We believe that these two concepts go together nicely.

ABSTRACT. This paper looks beyond of the current focus of research on biologically inspired cognitive systems and considers the problem of replication of its learned functionality. The considered challenge is to replicate the learned knowledge such that uniqueness of the internal symbolic representations is guaranteed. This article takes a neurological argument \no two brains are alike" and suggests an architecture for mapping a content of the trained associative memory built using principles of hyperdimensional computing and Vector Symbolic Architectures into a new and orthogonal basis of atomic symbols. This is done with the help of computations on cellular automata. The results of this article open a way towards a secure usage of cognitive architectures in a variety of practical application domains.

ABSTRACT. A “conscious system that can venture into an unknown environment” has been proposed. This study models the process of consciousness of a person who is going into an unknown environment. First, we assumed that to go into an unknown environment, the person needs to be curious about that environment and assured of its safety. Curiosity is a tendency to become interested in unknown phenomena and draw information from them. We consider that to acquire the behavior of going into an unknown environment (curiosity behavior), firstly the person needs in some way to go through many experiences of pleasure in unknown environments and increase curiosity and interest in such environments. To enter an unknown environment the person must also be assured that the environment is safe. We have developed a conscious system that can venture into an unknown environment and tested whether a robot can voluntarily enter an unknown environment.

ABSTRACT. There were many modeling studies on the emotion. But most of them were phenomenological and don’t approach to the brain or cognitive mechanism. In this study, we discuss on a possibility of its computational modeling based on an idea that an emotion in a wider sense is a value calculation system for an action decision. First, we show a possible brain architecture that the emotion system has a tight interaction with the cortical intelligent system through an interaction between the brain information processing areas of neocortex and limbic system. Then, in this study, we focused on a mechanism of value based intuitive inference on a path finding task in a partially known world. In a computer simulation, we tried to make a model of brainstem that is one of the brain emotion system parts. Though the brainstem looks simple, the model includes an essence of conflict resolving between multiple values.

ABSTRACT. Whole Brain Connectomic Architecture (WBCA) is defined as a software architecture of the artificial intelligence (AI) computing platform which consists of empirical neural circuit information in the entire brain. It is constructed with the aim of developing a general-purpose biologically plausible AI to exert brain-like multiple cognitive functions and behaviors in a computational system. We have developed and implemented several functional machine learning modules, based on open mouse connectomic information, which correspond to specific brain regions. WBCA can accelerate efficient engineering development of the intelligent machines built on the architecture of the biological nervous system.

ABSTRACT. While there are many types of cognitive architectures available today, one thing common to all of them is the need to cover the maximum number of the human intelligence factors used for solving various tasks. However, the currently existing cognitive architectures were developed based on a variety of aspects other than the human intelligence factors. One famous model of human intelligence factors is the CHC model which is studied in psychology. When it is used as the basis of a new cognitive architecture, the architecture will cover all the known human intelligent factors used for solving various tasks. In this paper, we propose a new cognitive architecture for dialogue situations based on the CHC model as the first step towards the formation of a comprehensive cognitive architecture. We will also outline the initial architecture using a case study.

ABSTRACT. Here we present the model of bio-inspired neuron, and synaptic plasticity, incorporating cellular homeostasis. Network of such neurons is used for multi-goal oriented control task. It was showed that such a model provides adaptive and robust behavior for the controlled agent.

ABSTRACT. This work aims at demonstrating that the neuromodulatory mechanisms that control the emotional states of mammals (specifically rat’s brains) can be represented and re-implemented in a computational model processed by a machine. In particular we specifically focus on two neurotransmitters, sero- tonin and dopamine, starting from their fundamental role in basic cognitive processes. In our specific implementation, we represent the simulation of the ‘disgust-like’ state based on the three dimensional neuromodulatory model of affects or emotions, according to the ‘cube of emotions’ where dopamine controls attention and serotonin is the key for inhibition. These functional mechanisms can be transferred into an artificial cognitive system: inhibition, for example, can elicit a blocking behaviour that, depending on its intensity and duration, can push the system to a general emotional state. Therefore, the main goal of this paper is to implement such a mechanism in a com- putational system to make it capable of managing a “failure” scenario in the complex set of inbound parameters appropriate for social environment useful for highlighting memories, decision making, resources evaluation, and other cognitive processes. We have simulated 1000 milliseconds of the sero- tonin and dopamine systems using NEST Neural Simulation Tool with the rat brain as the model to artificially reproduce this mechanism on a computational system. The results of the simulation experiments demonstrate the effectiveness of the proposed approach, pushing towards the completion of the biomimetic model by adding the third neurotransmitter (noradrenaline) and combining it with synthetic hormones.

ABSTRACT. This work presents the design, implementation and study of (1) a videogame -like virtual environment simulator, enabling social interaction of avatars controlled by human participants and by virtual actors; (2) a set of virtual actors with varying forms and degree of social-emotional intelligence, based on the eBICA cognitive architecture; and (3) a limited Turing test for social-emotional intelligence, involving human participants and virtual actors. The virtual environment simulator allows for various forms of emotionally-laden interaction of actors immersed in it in the form of avatars, with data collection characterizing their behavior in detail. The objective here is to compare and evaluate models of social -emotional reasoning based on the Turing test results and other objective behavioral measures, also taking into account subjective judgment of participants. One of the long -term goals is achieving human-level believability of socially-emotional virtual actors, such as non-player characters in games, personal assistants, robots, and other intelligent artifacts. Preliminary results indicate importance of social-emotional intelligence for believability, and support assumptions of the eBICA architecture.

ABSTRACT. This work continues the effort to design and test a universal cognitive model of emotionally biased behavior control and decision making, with the focus on social emotional relationships. Two key building blocks of the model include (i) dynamics of mutual appraisals of actors, determining the likelihoods of action selection, and (ii) moral schemas, or M-schemas, that establish normal behavior of two actors with respect to each other as well as to other entities, under certain implicit mutual relationships, such as partnership. To test the model, we implement a virtual actor embodied as an avatar in a specifically designed virtual environment, and use several paradigms of social interaction. Virtual environments and associated paradigms can be divided into a hierarchy, on top of which are paradigms with dynamically changing social relationships and roles. Using paradigms of this kind, we show that a virtual actor can be indistinguishable from a human participant, both, in its believability and in social acceptability; the latter being measured by the frequency of obtaining help from others. A model of this sort is expected to have broad applications in various fields in the near future.

ABSTRACT. The hippocampus has distinctive functional and structural properties. In this research, we utilize Restricted Boltzmann Machines (RBMs) based models inspired by distinctive structures found in hippocampus; neurogenesis in the dentate gyrus (DG) and recurrent connections in CA3. We review two types of topologically extended models of RBMs inspired by hippocampus. In one type of models, units are dynamically added during the training phase, and in the other type, connections are partly recursive. We analyzed these models both as separate models and combined model. The two types of the proposed models implement functions that the hippocampus has but the classical RBMs don’t. Furthermore, by combining the two proposed models, memorization of chronologically ordered data and memory reconstruction tasks’ performance improved significantly.

ABSTRACT. Reinforcement learning, which is a field of machine learning, is effective for behavior acquisition in robots. The extension of cognitive architecture with synchronous distributed systems is also effective for behavior acquisition. However, early work on the reinforcement learning software framework does not solve the difference between asynchrony and synchrony, and it cannot apply cognitive architecture with asynchronous distributed systems. Therefore, we applied asynchronous distributed systems to reinforcement learning modules to adapt to asynchrony and to extend cognitive architecture with asynchronous distributed systems. We prototyped a reinforcement learning software framework called “Re:ROS."

ABSTRACT. In this study, we clarify the integration mechanism of sensory information of vision and somatosensory sensation in walking. In this experiment, we evaluated the possibility of affecting walking by attenuating a somatosensory sensation by giving vibration stimulation to the feet, and by presenting optic flow to the peripheral visual field to generate a self - motion sensation of vision superiority. Experimental results confirmed that walking in the direction opposite to the self - motion sensation is presented by presenting the optic flow and vibration stimulation. Based on the results of this experiment, we propose that sensory devices such as vision and somatosensory sensation are not exclusive in walking, but are integrated by superposition.

ABSTRACT. We have been proposing a computational model of the cerebral cor- tex called BESOM, which models the cerebral cortex as restricted Bayesian net- works based on recent findings in the neuroscience area. Since BESOM is based on Bayesian network, it inherently allows bi-directional information flow, meaning that it can naturally merge information extracted from concrete data with highly- abstract prior knowledge. As an example of such kind of tasks, we report robust text recognition task with context information. We show that word spelling knowledge and word n-gram could be represented as a part of the network and actually they contribute the text recognition accuracy with noisy text images. We also show that the computational cost is approximately linear with the number of characters and words.

ABSTRACT. Recent advances in the face detection (FD) and recognition (FR) technology make it appear that the problem of face image localization and matching is essentially solved, e.g. via deep learning models using thousands of samples per face for training and validation on the available benchmark data-sets. Human vision system seems to handle face localization and matching problem differently from the modern FR systems, as humans appear to detect faces instantly even in most cluttered environments, which may be evolutionary hard-wired and experience-tuned, and often require a single view of a face to reliably distinguish it from all others. This prompted us to take a biologically inspired look at building a cognitive architecture that extensively uses artificial neural nets at the face detection stage and adapts a single image per person (SIPP) approach for face image matching.

ABSTRACT. We describe approach, architecture, implementation and practical applications of personal software agent with social awareness, capable to capture socio-temporal context of its user on the Web and in social networks in the course of interactions of the user with agent itself and user’s Internet environments online.

ABSTRACT. Paper presents a Modular Approach for Time series analysis area. We consider the most important characteristics of modular architectures of neural networks and their advantages under traditional monolithic neural networks. The main idea of this paper is take answer - why modular neural networks have so high performance in many tasks. Also we present few examples of modular approaches which can be applied for time series analysis problem.

ABSTRACT. What is “self-awareness”? How can explicit consciousness and sub-consciousness be mapped in relation to each other? How are they related to the self? How can these entities be represented in an artificial conscious system? These questions are the focus of this article. People are aware of only the behavior that they are focusing on; they cannot be directly aware of routine behavior such as walking and breathing. The latter is generally called unconscious behavior, and here we call it sub-conscious behavior. To understand self-awareness, therefore, firstly it is important to map explicit consciousness and sub-consciousness, which is where the self is deeply involved. We consider that if there is no self that refers to itself, no one can be aware of what he himself is doing. In this study we map explicit consciousness and sub-consciousness using an artificial conscious system, and then make a new proposal about the relationship between self-awareness and the self.

ABSTRACT. Deep learning has attracted significant interest currently. The deep Q-network (DQN) combined with Q-learning have demonstrated excellent results for several Atari 2600 games. In this paper, we propose an exploitation-oriented learning (XoL) method that incorporates deep learning to reduce the number of trial-and-error searches. We focus on a profit sharing (PS) method that is an XoL method and combine a DQN and PS. The proposed method DQNwithPS is compared to a DQN in Pong of Atari 2600 games. We demonstrate that the proposed DQNwithPS method can learn stably with fewer trial-and-error searches than only using a DQN.

ABSTRACT. The WBA approach is an approach to realize artificial general intelligence by ‘mimicking’ the architecture of the entire brain. It is argued that referring to the brain architecture could help to attain a unifying and generally accepted framework for the design, characterization, and implementation of human-level AGI, for the human brain is, of course, the organ that realizes human-level intelligence and its architecture can be used as a shared reference architecture. Having a reference architecture would facilitate collaboration among researchers and the realization of human-level AGI. The reference is now more realistic than before, as we have more neuroscientific knowledge of brain architecture (such as connectome) and functions and more practical and theoretical knowledge on information processing or machine learning to hypothesize the working of the brain. In this panel, participants will discuss how a community of researchers could work together to create a unifying and generally accepted framework by referring to the brain to realize human-level intelligence. Notably, connectomic architecture, the modeling of brain organs such as the neocortex, hippocampus, basal ganglia, amygdala, and cerebellum, and required learning algorithms to be shared will be discussed. Besides architecture, tools for agent simulation and for neuroinformatics to be shared for research would be discussed. If time permits, a roadmap with shared frameworks and tools may be discussed. Researchers with ideas on this topic are encouraged to be discussants. The format of the discussion will be round-table.