ABSTRACT. We present a computational model that describes the early stages of visual processing and the within selective attention mechanisms to generate feature-based (hue or color) activations of salient localizations based on neurophysiology evidence of selective responses in the visual pathway. The model identifies related brain areas, the feasible computations of each one, and proposes the type of data generated and shared among the components. This work is part of the selective aspect of an attention system designed for a broader cognitive architecture for virtual creatures.

ABSTRACT. Single-shot grid-based path finding is an important problem with the applications in robotics, video games etc. Typically in AI community heuristic search methods (based on A* and its variations) are used to solve it. In this work we present the results of preliminary studies on how neural networks can be utilized to path planning on square grids, e.g. how well they can cope with path finding tasks by themselves as well as how they can be put into service to the conventional heuristic search algorithms, e.g. A*. Conducted experiments show that the agent robustly learns to achieve the goal and that the adaptive heuristic is capable of reducing breadth of the search tree.

ABSTRACT. The ability to infer goals, consequences of one’s own and others’ actions is a critical desirable feature for robots to truly become our companions-thereby opening up applications in several domains. This article proposes the viewpoint that the ability to remember our own past experiences based on present context enables us to infer future consequences of both our actions/goals and observed actions/goals of the other (by analogy). In this context, a biomimetic episodic memory architecture to encode diverse learning experiences of iCub humanoid is presented. The critical feature is that partial cues from the present environment like objects perceived or observed actions of a human triggers a recall of context relevant past experiences thereby enabling the robot to infer rewarding future states and engage in cooperative goal-oriented behaviors. An assembly task jointly done by human and the iCub humanoid is used to illustrate the framework. Link between the proposed framework and emerging results from neurosciences related to shared cortical basis for ‘remembering, imagining and perspective taking’ is discussed.

ABSTRACT. The paper considers theoretical mechanical model of a multifingered arm with 21 degrees of freedom. The main objective of the work - is the synthesis of finger control schemes in the tasks of collaborative robotics, as well as gesture recognition task with the help of neural network training. As the demonstration we propose to observe the results of three gestures recognition with the help of constructed convolutional network. For three lables (classes) 201 images at different distance and at different angles were created. As a result of neural network training the accuracy of classification is 67 percent.

ABSTRACT. A modelling and simulation approach is discussed for nonlinear controlled dynamical systems under multiple and diverse uncertainties. The main goal is to demonstrate capabilities for semi-empirical neural network based models combining theoretical domain-specific knowledge with training tools of artificial neural network field. Training of the dynamical neural network model for multi-step ahead prediction is performed in a sequential fashion. Computational experiments are carried out to confirm efficiency of the proposed approach.

ABSTRACT. A simulation problem is discussed for nonlinear controlled dynamical systems represented by differential-algebraic equations of index 2.The problem is proposed to be solved in the framework of the semi-empirical approach combining theoretical knowledge for the plant with training tools of artificial neural network field. Special form network based semi-empirical models implementing implicit Runge-Kutta method of numerical integration are proposed to use. The training of the semi-empirical model allows to elaborate the models of aerodynamic coefficient implemented as a part it. The results of simulation for elaboration procedure of lift coefficient in respect to reentry hypersonic vehicle are presented.

ABSTRACT. Standard computing will be characterized as a functional extension of syntax. This is based, partly, on Searle’s Chinese Room argument. Using BICA philosophy, in particular the claim of continuity between human-animal-robot cognitive architectures, we will define strong semantics as the ability of a cognitive architectures to consult cognitive maps, in particular phenomenal content map. The goal of strong semantic computing in autonomous robotics should be to ‘know what is going on’ before engaging in detailed logical analysis (it can be called Gestalt computing). Such computing is needed in advanced autonomous robotics, especially robots functioning in human environments. Incidentally, this approach provides a partial solution to Searle’s Chinese room case.

ABSTRACT. Deep learning models of vision have recently achieved human-level of performance in tasks such as object recognition and face identification. However, the extent to which these computational models resemble the way biological vision works remains unclear. We address this question by investigating how a deep learning model of face identification represents faces and by comparing the results to data from behavioural experiments. To this end, we use the Bubbles technique (Gosselin & Schyns, 2001) in order to find the critical regions within an image which are needed for correct face identification. The technique is applied to a pretrained deep convolutional neural network model of face recognition (Amos, Ludwiczuk, & Satyanarayanan, 2016), and the results are related to human data from a replication of Gosselin & Schyns (2001). We also use Bubbles to check how faces are represented in a simple connectionist model and in a deep learning model trained on general object recognition.

ABSTRACT. In this paper the cognitive architecture based on the Functional Systems Theory (TFS) by P.K.Anokhin is presented. This architecture based on the main notions of this theory: goal, result, anticipation of the result. This theory is described on physiological and informational level. The logical structure of this theory was analyzed and used for the control system of the purposeful behavior development. This control system contains the hierarchy of functional systems that organize the purposeful behavior. The control system was used for the agents modeling that are solving the foraging task. The computer experiments are presented that compare this control system with the control systems based on the reinforcement learning.

ABSTRACT. Emotional and motivational evaluations are part of the development of rewards within living beings. Particularly, during the perception of stimuli in the environment, these evaluations collaborate with one another to generate reward values automatically, without the need to involve rational processes. In this paper we propose a conceptual model of automatic reward for virtual creatures inspired by neuroscientific evidence, contemplating the processes of emotions and motivations, as well as the generation and recovery of automatic reward values. According to the evidence, the reward process is divided into two processes: liking, which is oriented toward interpreting information inputs and generating reward values, and wanting, focused on the recovery of the stored reward values and the generation of objectives in the environment. The reward process is implemented as a concurrent and parallel naturally distributed system, allowing virtual creatures to adapt to their environment and generate more credible behaviors. The results of liking and wanting are shown in this article through a case study, in which the performance of both processes is observed when the creature interacts with the environment.

ABSTRACT. Nowadays the number of huge companies and corporations has in their disposition various non-structured texts, documents and other data. The absence of clearly defined structure of the data makes the implementation of searching queries complicated and even impossible depending on the storage size. The other problem connected with staff, which may face the problem with misunderstanding of the special query languages, knowledge of which is necessary for the execution of searching queries. To solve these problems, we propose the semantic search system, the possibilities of which include the searching index construction, for queries execution and the semantic map, which would help to clarify the queries. In this paper we are going to describe our algorithms and the architecture of the system, and also to give a comparison to analogues.

ABSTRACT. The performance of the human brain depends on how effectively its distinct regions communicate, especially the regions which are more strongly connected to each other than to other regions, or so called “rich-clubs”. The aim of the current work is to find a connectivity pattern between the three brain rich-club regions without any a priori assumptions on the underlying network architecture. Rich-clubs for the analysis were previously identified with structural MRI. Functional magnetic resonance imaging (fMRI) data from 25 healthy subjects (1000 time points from each one) was acquired and Transfer Entropy (TE) between fMRI time-series from rich-clubs was calculated. The significant results at the group level were obtained by testing against the surrogate data generated on a novel approach. We found stable causal interactions between rostral Anterior Cingulate Cortex L and Dorsal Anterior Cingulate Cortex L, dorsal Anterior Cingulate Cortex L and Paracentral Lobule R but not vice versa. Our work provides an approach to causal analysis of experimental data and demonstrates the applicability to real fMRI study.

ABSTRACT. Dynamic model of a recurrent neuron with sigmoidal activation function is considered. It is shown that the neuron activation characteristic (dependence between an input pattern and an output signal) can have the form of both a smooth sigmoidal function and a step function in the form of a quasi-rectangular hysteresis loop due to the presence of the modulation parameter. We demonstrate that a gradual behaviour in the network can be implemented which means that neurons will have output values proportional to how each of them corresponds to the input pattern. In this case signals in the network will be propagated freely but with attenuation. The output values of the neurons also will be low. At the same time a winner-takes-all behaviour can be implemented which means that neurons becomes similar to threshold units. As a result, there will be limited network activity since only a part of the neurons will have output values close to the maximum value while the remaining neurons will have output values close to the minimum value. Thus the presence of the modulation parameter provides a means to change quickly the behavior strategy of the network which affects the processes of pattern recognition and learning.

ABSTRACT. We propose a molecular associative memory model, by combining auto-logistic specifications which capture statistical dependencies within the local neighborhood systems of the exposed knowledge, with the bio-inspired DNA-based molecular operations which store and evolve the memory. Our model, characterized by only the local dependencies of the spatial binary data, allows to capture only a fewer features. Our memory model stores the exposed patterns and recalls the stored patterns through bio-inspired molecular operations. Our molecular simulation exemplifies the applications of associative memories in pattern storage and retrieval with high recall accuracy, even with lower order memory traces (pair-wise cliques) and thus exhibits brain-like content-addressing cognitive abilities.

ABSTRACT. Biomedical signals such as electroencephalogram (EEG) are the time varying signal, and different position of electrodes give different time varying signals. There might be a correlation between these signals. It is likely that the correlation is related to the actual position of electrodes. In this paper, we show that correlation is related to the physical distance between electrodes as measured. This finding is independent of participants and brain hemisphere. Our results indicate that the EEG signal is not transmitted via neurons but through white matter in a brain.

ABSTRACT. In this paper, we present an overview of MECA, the Multipurpose Enhanced Cognitive Architecture, a cognitive architecture developed by our research group and implemented in the Java language. MECA was designed based on many ideas coming from Dual Process Theory, Dynamic Subsumption, Conceptual Spaces and Grounded Cognition, and constructed using CST, a toolkit for the construction of cognitive architectures in Java, also developed by our group. Basically MECA promotes an hybridism of SOAR, used to implement rule-based processing and space-state exploration in System 2 modules, with a Dynamic Subsumption Motivational System performing the role of System 1, using a representational system based on conceptual spaces and grounded cognition. We review the conceptual background used on MECA and further provide a detailed description of the many MECA sub-systems.

ABSTRACT. This paper outlines an augmented robotic architecture to study the conditions of successful Human-Humanoid Interaction (HHI). The architecture is designed as a testable model generator for interaction centred on the ability to emit, display and detect honest signals. First we overview the biological theory in which the concept of honest signals has been put forward in order to assess its explanatory power. We reconstruct the application of the concept of honest signalling in accounting for interaction in strategic contexts and in laying bare the foundation for an automated social metrics. We describe the modules of the architecture, which is intended to implement the concept of honest signalling in connection with a refinement provided by delivering the sense of co-presence in a shared environment. Finally, an analysis of Honest Signals, in term of body postures, exhibited by participants during the preliminary experiment with the Geminoid Hi-1 is provided.

ABSTRACT. We present an online system for real time recognition of actions involving objects working in online mode. The system merges two streams of information pro- cessing running in parallel. One is carried out by a hierarchical self-organizing map (SOM) system that recognizes the performed actions by analysing the spa- tial trajectories of the agent’s movements. It consists of two layers of SOMs and a custom made supervised neural network. The activation sequences in the first layer SOM represent the sequences of significant postures of the agent during the performance of actions. These activation sequences are subsequently recoded and clustered in the second layer SOM, and then labeled by the ac- tivity in the third layer custom made supervised neural network. The second information processing stream is carried out by a second system that determines which object among several in the agent’s vicinity the action is applied to. This is achieved by applying a proximity measure. The presented method combines the two information processing streams to determine what action the agent per- formed and on what object. The action recognition system has been tested with excellent performance.