The programme is designed to exchange the results of solving Data Science experiments, assess the qualitative characteristics of these results, and upload the most promising solutions to the production server environment. It consists of API managing file data storage and MLOps-system, as well as a database for storing solutions offered by users.
Computer type: PC based on Intel x86 processor
OS: CentOS-7
The programme is designed to generate synthetic tabular data, time series and images, includes implementation and evaluation of regression and classification algorithms, assessing their performance on both generated and raw data. Functionality: loading of training data, analysis of unique values, generation of synthetic data using generative artificial intelligence (AI) models WGAN, VAE, Real NVP, use of metrics to measure the level of confidentiality and quality of generated synthetic data, their difference from real data, visualisation of research results.
Программа предназначена для процесса обучения символьной регрессии на основе автокодировщика. Обеспечивает устойчивость к шумам, возможность настройки априорных значений для символьного выражения для применений физики и др.
The program is designed for modelling radio wave propagation by ray tracing between a transmitter and a moving receiver, calculating signal quality indicators, collecting and processing the obtained data. The program contains libraries for ray tracing modelling, working with arrays and tables. The program includes a module for modelling radio waves with ray tracing, calculation of modelling indicators, saving of selected indicators, module for processing of obtained data.
Computer type: IBM PC - joint.PC based on 10*3.2 GHz processor.
OS: MAC OS 12.6
The software provides the user with the ability to convert ray tracing data into a frame sequence format, configure and train a neural network based on it and then save it.
The programme is designed to classify the states of an industrial three-phase motor by methods of machine learning neural networks on the data of the current signal of the motor supply, defined as: 1) normal state, corresponding to a serviceable motor; 2) motor inter-turn faults; 3) mechanical defects of the motor. The programme is based on the architecture of a one-dimensional convolutional neural network consisting of 14 layers and 26 thousand parameters. The accuracy of classification of motor states on test data exceeds more than 2 times the accuracy according to GOST ISO 20958-2015.
Computer type: HP ProLiant DL-160 G6 server or analogue
OS: Ubuntu 14.04.5 LTS (Trusty Tahr) and higher.
The programme is designed to store terms and vector representations, exchange and manipulate data for semantic analysis and was developed due to the need to extract vectors and related features from heavyweight machine and deep learning models.
Computer type: HP ProLiant DL-160 G6 server or equivalent
OS: Ubuntu 14.04.5 LTS (Trusty Tahr) or higher
The program allows to solve the tasks of standardisation of data from different information systems to ensure quality, reliability and validity, the task of ensuring transparency of data acquisition by storing references to their sources and combining named entities and subject terms, search for terms in Russian and English. The application area of the programme is linguistic analysis of texts in Russian and English.
Computer type: HP ProLiant DL-160 G6 server or analogue
OS: Ubuntu 14.04.5 LTS (Trusty Tahr) and higher
The programme allows to predict the volume of the goods market by the textual following of the corresponding terms and is applicable for scientific and technological forecasting in the interests of strategic analytics and foresight studies.
Computer type: HP ProLiant DL-160 G6 server or analogue
The programme checks facts in several languages (Russian, English, Bulgarian) based on sources from Wikipedia. For each language the local version of Wikipedia is used. The input data is a set of several facts entered by the tool user. The output is the result of checking each statement, represented by one of three answers: 1) Truth; 2) Lying; 3) Not Enough Info.
The programme is based on keyword extraction. The algorithms are designed in such a way that the most relevant words and phrases are extracted as a result of text processing. The programme receives arrays of unstructured text data as input. The text of the document is broken down into sentences, dependencies between words are identified based on analysing the syntax of the processed sentence. Based on the identified dependencies between words, individual nouns and adjectives are combined into word combinations, n-grams, to which the NER model assigns tags, which are indications of named entities. By means of ranking implemented on the basis of selected TF-IDF, RAKE, BM25, KeyBERT algorithms, key metainformation is selected from the whole array of available metainformation. The programme is applicable for a wide range of tasks of semantic analysis and information retrieval.
Computer type: HP ProLiant DL-160 G6 server or analogue Ubuntu 14.04.5 LTS (Trusty Tahr) or higher
The programme is designed to automate the process of creating models for detecting objects and actions. Functionality includes: preprocessing of dataset, including images and annotations; generation of configuration files; auto selection of hyperparameters; selection of neural network architecture; early stopping algorithm with saving the best checkpoints, functionality for detecting correctness of arbitrary sequence of objects.
Computer type: IBM PC - joint. PC
OS: Linux, MacOS, Windows
The software allows analysing video stream from high-resolution cameras using computer vision technologies and thus controlling productivity and workplace safety when performing manual operations in production. Algorithms draw conclusions about the errors made and the stages of the assembly process passed. Flexible configuration of the application allows to adapt it to a new technological process in a short time.
Computer type: IBM PC-compatible PC
OS: Ubuntu 22.04
The programme is designed for automatic preparation and markup of a data set for further training of neural network models designed to solve the detection problem. The program is customisable and can create datasets in which each photo contains objects of different classes with specified distributions of their number and location.
OS: Ubuntu 22.04
The invention relates to the field of computer technology for quality control of assembly products. The technical result is to improve the quality of assembly by controlling the correctness of the final set of regulated assembly operations.
The programme consists of three modules that train and test different graph networks based on the prepared data and draw additional graphs (ROC-, PR- and F1 score curves). As input, the programme receives the organism's genome, as well as the results of chip-atlas experiments converted into sparse vector format. It is also necessary to manually specify the model parameters and the balance of classes of the training dataset. The model is trained on a training sample (first module) and then tested on the full genome (second module). In the third module, ROC-, PR- and F1 score curves can be plotted on the required dataset.
Computer type: IBM PC-compatible. PC.
OS: Mac OS X 10.11 or higher, Linux Ubuntu 16.04 or higher
The framework integrates software modules for predicting the location of genomic functional elements (GFEs) using deep learning methods, allowing to select the type, architecture of neural networks for predicting the location of GFEs and omics data, to complement with transfer learning methods, to select prediction parameters and to evaluate its quality.
Computer type: IBM PC - joint
PC. OS: Mac OS X 10.11 or higher, Linux Ubuntu 16.04 or higher
The programme reulises full-genome analysis using the DNABERT transducer algorithm trained on experimentally identified sequences forming Z-DNA (Z-flipons). The algorithm provides a significant performance improvement (F1 = 0.83) over existing approaches and implements computational mutagenesis to assess the effect of base substitutions on Z-DNA formation.
The programme provides a graphical interface for applying a pre-trained machine learning model (multilayer perceptron) to assess the presence and degree of dyslexia in a schoolchild based on gender, age, school grade and oculography data. To use the programme, you need to specify the patient's demographic data and attach a .CSV file with oculography data (duration and pupil fixation parameters), the result is belonging to one of three classes - dyslexia, dyslexia risk or normal.
The first artificial intelligence-based solution with robust results for detecting dyslexia from eye movement data of Russian-speaking schoolchildren in grades one to six. Comprehensive study of the performance and hyperparameter fine-tuning of ten classification and seven regression algorithms. Proposed the second largest eye movement data set in this area of research, consisting of three discrete target values and one continuous one.
The program is designed to clarify the origin of a human being by analysing the commonality of his genome segments with representatives of reference populations. It is applicable for increasing the informativeness of commercial genetic tests. It contains a set of graph neural network architectures that have shown the best classification quality on model data and heuristic classifiers for comparative analysis. It allows to train neural networks on marked-up data and to predict the population affiliation of new individuals for whose genome IBD segments with individuals in marked-up data are defined.
The programme integrates modules for data processing, neural network architecture, training and testing, and a module for generating artificial genotypes containing epistasis. The input set of genetic data in vcf format is obtained as a result of chip sequencing or generation by the corresponding module. In the data preparation step, the most significant single nucleotide polymorphisms are highlighted, selected based on published GWAS studies. The models estimate the risk of disease (in %), the quality of prediction is evaluated using a number of metrics (ROC AUC, PR AUC, F1, precision and recall).
Computer type: IBM-PC compatible PC.
OS: Mac OS X 10.11 or higher, Linux Ubuntu 16.04 or higher.
The programme is a cross-platform application aimed at correcting morphosyntax (agrammatism) disorders in adults with speech disorders (aphasia) of various etiologies. In the course of training, lexical access, sentence structure, person/number in the present tense, gender/number in the past tense, verb tense, verb control, prepositional control, declension of adjectives are practised.
Computer type: IBM PC-compatible. PC
OS: Android, iOS
The environmental monitoring system reduces the time and increases the accuracy of predicting the spatial distribution of harmful substances in the atmospheric air, and has the ability to simultaneously cover a plurality of local, located at a single industrial site, and a plurality of territorial, connecting a plurality of local environmental monitoring systems of different industrial sites. The invention can be used for integrated planning and notification of risks of atmospheric air pollution of industrial enterprises by harmful substances.
The programme is designed to predict meteorological variables (temperature, precipitation, pressure, humidity, wind speed) using WRF (Weather Research and Forecasting) simulator. The programme consists of two modules. The first module generates the training sample. The second module is responsible for optimising the WRF hyperparameters.
Computer type: Intel processor based PC
OS: Fedora Linux 36
The programme is designed to predict meteorological variables (temperature, precipitation, pressure, humidity, wind speed) using WRF (Weather Research and Forecasting) simulator. The programme consists of two modules. The first module generates the training sample. The second module is responsible for optimising the WRF hyperparameters.
Computer type: Intel processor based PC
OS: Fedora Linux 36
The invention relates to the field of environmental monitoring and can be used to detect sources of atmospheric air pollution at industrial enterprises. Essence: continuous measurements of concentrations of harmful substances in the atmosphere are performed in real time using monitoring stations. Besides, with the help of meteorological stations continuous measurements of wind speed and direction are performed. The results of the measurements are fed into the main computer, the central processor device of which processes the information obtained with the help of an intelligent analytical system - an artificial intelligence model. In case of registration of events of unauthorised emission of harmful substances in the atmospheric air with the help of the artificial intelligence model in accordance with the trained search strategy determine the estimated locations of the source of atmospheric pollution on the marked map of the industrial area by moving the search cursor. The nearest source of air pollution to the estimated location is identified. Technical result: reduction of time of detection of the source of unauthorised emission of polluting harmful substances into the atmosphere.
The utility model relates to control and measuring devices for ecological monitoring of urban environment and is intended for processing and control of measurement results of concentration of pollutants of various types in the atmospheric air, as well as ensuring the operability of sensors measuring the concentration of pollutants of various types in the atmospheric air with the help of artificial intelligence.
The programme consists of a number of interrelated modules: a module for collecting and processing text materials, a module for classifying text materials by sentiment using artificial intelligence, and a module for calculating various sentiment metrics. It provides automated collection, processing and detection of sentiment (tone) of investment-themed text materials on Telegram channels.
The programme allows solving classification and regression problems in financial economics using dense layer neural network models and interpreting the results by means of explanatory artificial intelligence (AI) (Shepley vectors), for example, to identify determinants of stock and bond exchange characteristics. The programme includes modules for data loading and unloading, building neural network models for regression and classification tasks, explanatory AI and graph generation.
Computer type: IBM PC-compatible. PC
OS: Windows 7/8.1/10/11
The utility model relates to accessories for smartphones and can be used as a security device including additional functionality in the form of support for investment decisions based on artificial intelligence.
The programme implements a predictive model of artificial intelligence (AI) based on the architecture of transformer class with wavelet embedding. The distinctive features of AI models of this class are the presence of encoder and decoder blocks, as well as the attention mechanism. One of the directions of use is forecasting of time series in the field of financial economics. The programme consists of interconnected modules: wavelet embedding of the initial time series of financial data; encoder; attention mechanism; decoder; experiment execution/data loader.
Computer type: IBM PC-compatible. PC
OS: Windows 7/8.1/10/11
The programme is designed to determine the optimal average tariff for different hotel chains/individual hotels. The programme can be used by the management staff of hotel chains/individual hotels, including for setting pricing policy. The functionality of the programme includes cost determination by optimising the revenue of hotel chains/individual hotels.
Computer type: IBM PC-compatible. PC
OS: Windows 2000, XP, NT and Mac OS.
The programme solves the problem of automatic assignment of points for open-ended tasks of a digital tool for assessing the reading literacy of 3rd grade students. Three classification models are developed for each of the three tasks of the tool. The programme can be used by developers of open-type tasks and experts checking such tasks.
Computer type: IBM PC-compatible. PC
OS: Windows 7/8/8/8.1/10, Linux
The programme is designed to determine the psycho-emotional state of online lecture students using real-time video images of their faces. The method of determining the emotional state of online lecture participants and their involvement is implemented using lightweight neural networks from the HSEmotion library, as well as the face tracking mechanism.
Computer type: IBM PC-compatible. PC
OS: Windows, Linux, macOS
The programme is a web application that allows to determine the ethical index of companies for different sectors of the economy. The programme automatically collects data, including text data, from public sources and saves it into a database. Based on the obtained data, an ethics index is calculated using artificial intelligence models, which can then be visualised for the user.
Computer type: IBM PC-compatible. PC
OS: Windows 7/8/8/8.1/10; Ubuntu 18.04/20.04/22.04
Forecasting the number of cases of arbitration courts of the Russian Federation
More
The program is designed to generate a forecast of the number of arbitration cases by category for future time periods based on the analysis of data accumulated in the data warehouse. The programme can visualise the number of existing data in the system in the form of graphs.
Computer type: IBM PC compatible computer, 4x2.5 GHz processor
OS: Windows 11
The programme is designed to find out the entropy structure of time series describing the number of new cases in the RF court. The program can be used for automation and improvement of quality of work of courts of the Russian Federation, departments of statistics and scientific divisions of legal centres. Main external functions: input of data on the number of new cases in the court for a certain period, region, category; obtaining of describing statistics; conclusion on the structure of the series.