Annotation: The paper presents examples of modern large-scale linguistic models. The implementation problems of intelligent robot human-machine interface as well as their solutions are discussed. Finally, the algorithms of natural language human-machine interface intelligent robot in natural language are proposed.

Annotation: The problem of finding approaches to creating data sets based on features extracted from time series recorded when working with the 14-channel neural interface EMOTIV EPOC+ in the process of formation of motor imageries by independent subjects for controlling the computer mouse cursor is considered. The purpose of the work is to study a number of indicators, the values of which are calculated by performing certain mathematical operations on various fragments of 14 time series, as tools for generating feature values. The study should help to identify those indicators, the use of which to form features in datasets ensures the development of motor imagery classifiers that differ in the highest possible quality of data classification. Previously, the time series were filtered using a 5th order Butterworth filter, which made it possible to solve the problem of removing noise artifacts from the time series. To assess the influence of the length of a fragment of a time series used to form the value of a feature based on a particular indicator on the final quality of data classification, time intervals (time frames) of 1, 2 and 3 seconds were considered. During the research, the development of SVM (Support Vector Machine), RF (Random Forest) and MLP (Multi Layer Perceptrone) classifiers of motor imageries was carried out. The experimental results showed the feasibility of working with indicators calculated based on Shannon entropy and Higuchi fractal dimension on a time frame of 3 seconds. In this case, it is possible to ensure high quality of classification of motor imageries, assessed using the F-measure. In particular, SVM, RF and MLP classifiers developed on the basis of datasets in which features are calculated using Shannon entropy have the maximum Fmeasure values. These F-measure values are 0.82, 0.88 and 0.73, respectively.

Annotation: The article presents a method for segmenting a point cloud of a scene consisting of unknown objects based on the use of the Voxel Cloud Connectivity Segmentation (VCCS) method and two-stage feature vector processing using the PointNet neural network and Dynamic Graph Convolutional Neural Network (DGCNN). In some cases, the practical application of manipulative robots involves grasping objects whose shape, color and other features are not known in advance. In particular, examples of such tasks can be cleaning of premises, emergency rescue operations to remove blokage, work in warehouses or in post offices. In the proposed approach, an image of the scene in the form of a point cloud is compiled from a set of images of a cluttered scene obtained from RGBD cameras, then this point cloud is processed using the VCCS heuristic algorithm and machine learning methods. The result of the approach is a segmented point cloud, for each point of which there is a label that determines its belonging to a separate object in the scene. The novelty of the approach lies in the combination of the VCCS heuristic algorithm and the new neural network architecture, which is a combination of modified PointNet and DGCNN networks. The conducted experimental studies confirm the operability of the proposed solution.

Annotation: This article proposes a new approach to solving the clustering problem with cutting off outliers, uninformative anomalous data and other information noise for streaming data in the feature space of any dimension and with the memory of all processed data points. To implement this task, an original modification of the DBSCAN algorithm was developed, using a hybrid approach to finding the boundaries of clusters of arbitrary shape and determining whether each of the data points is located inside or outside such a boundary. During the development, both machine learning technologies and mathematical methods were used, in particular, the method of calculating the convex hull of a finite set of points in the n-dimensional Quickhull space. The resulting algorithm consists of several blocks that are activated depending on the nature of the distribution of data received from the input stream. The application of the developed algorithm guarantees the creation of a closed cluster boundary of arbitrary shape. Using the adaptive frame splitting mechanism, it allows clustering of data of different dimensions and large volumes, with the memory of all incoming points.As a result, the authors managed to create a modification of the DBSCAN algorithm for streaming data that is efficient in terms of execution speed and memory usage. To illustrate the efficiency, gain of the developed algorithm modification in comparison with the classic DBSCAN variant, a calculated assessment of performance and memory requirements was carried out. The correctness of the estimates obtained has been confirmed experimentally. The presented modification of the DBSCAN algorithm for streaming data not only is able to get an overall performance gain with lower memory requirements compared to the classic DBSCAN algorithm, but also has functional advantages associated with the ability to work efficiently with streaming data in the presence of information noise. These advantages make the presented modification of the DBSCAN algorithm useful for solving complex problems in streaming data processing systems, such as searching for correlations and anomalies in statistical indicators of distributed data collection systems or for detecting stable states of queuing models used in logistics and transport.

Annotation: The article describes the creation of models for predicting the probability of students based on the analysis of data on the vulnerability of measuring materials and class attendance. During the study, methods of correlation analysis and machine learning were used, as a result of which a linear discriminant analysis (LDA) model was selected, which showed good results. The developed model can help improve the quality of education, and can also be adapted for use in other areas of forecasting based on multiple processes.

Annotation: The urgency of the need to develop models and research methods for bot-cyber-attacks as IaaS-class cloud systems used externally for quite legal purposes is shown. The results of the analysis of the structure of botnets, cyberattacks, and models of computational substantiation of resource provision and deployment, both IaaS and IaaS services it was determined and formalized scientific problem of estimating the number of virtual connections IaaS in the interests of monitoring and identification of botnets, cyber-attacks, the results of which received a new optimization model of the system "sources of information different categories of content – fragment IaaS" . The method of covering areas has been developed for calculating the total number of IaaS virtual connections. The obtained scientific results can be used as the basis for automating the computational justification of multi-variant IaaS design solutions at the pre-project stage.