OCR for Arabian documents for carsharing: - detection and recognition of significant fields on different documents; - methods of smart data synthesis in situations of their acute shortage and very distorted truth. Распознавание текста на арабских документах для задач автоматического каршеринга: - поиск и распознавание значимых полей на различных документах; - методы тонкого синтезирования данных в ситуации, когда реальных данных мало, а truth очень загрязнен. (Tensorflow detection API , Keras OCR fine tuning, Keras with GRU and CTC-loss, VGG, SSD and MobileNet, cv2 and computer vision metods)

Automatic data collection about road objects for self driving tasks: - Traffic signs localization and classification on the series of consecutive frames. Data generation methods in situations with sharply unbalanced dataset; - road marking detection and road segmentation; Автоматический сбор данных о дорожных объектах для составления высокоточной дорожной карты в проекте беспилотного вождения. - обнаружение и классификация дорожных знаков по серии последовательных снимков; - генерация данных в ситуации сильной несбалансированности классов в данных; - обнаружение дорожной разметки и границ дорожного полотна. (Tensorflow detection API, Keras, LeNet, VGG, SSD and MobileNet, scnn-tensorflow, pyTorch, cv2 and computer vision);

License Plate recognition Camera Systems - data mining, feature engineering, machine learning: - Detection, segmentation and symbols recognition with convolutional neural networks (Tensorflow detection API, Keras, LeNet, VGG, SSD and MobileNet) ; - image comparison and duplicate search (siamese net); - Merging outcomes from several recognizers (feature engineering, XGBoost); - Image processing, image feature engineering (python, OpenCV). Участие в разработке системы автоматического чтения автомобильного номера, встроенной в дорожную камеру - анализ данных, подготовка факторов, машинное обучение: - Локация, сегментация и чтение автомобильного номера с помощью глубоких нейронных сетей (Tensorflow detection API, Keras, LeNet, VGG, SSD, MobileNet) ; - Сравнение изображений и поиск дубликатов (siamese net); - Слияние результатов работы нескольких распознавателей (feature engineering, XGBoost); - Алгоритмы обработки изображений, подготовка факторов (python, OpenCV).

Natural Language Processing, feature engineering, machine learning, development (MatrixNet, word2vec). Taking part in projects of Yandex Search system: – automatic correction of typos; – Automatic grouping queries; – Automatic search for semantically related queries; – spam detection; – Automatic selection of video and web queries. Обработка естественного языка, разработка признаков, машинное обучение, разработка. Работа в поисковом сервисе: - автоматическое исправление опечаток; - Автоматическая группировка запросов; - Автоматическое выделение связанных по смыслу запросов; - Распознавание "накрученных" и "роботных" запросов; - Автоматическое разделение запросов по темам.

Working in all steps of the recognition process: location, noise elimination, segmentation, extraction of the significant features for recognition, data analysis, training of neural networks, merging the results. Taking part in the following projects of the company: – automatic search of breast cancer on a mammogram and tomogram (tomosynthesis) from detection to verification and evaluation of malignancy (7 years); – automatic recognition of the complete U.S. mailing address, including all types of address data: machine print, handprint and cursive; – automatic recognition of check courtesy and legal amount in check images; – detecting signature presence and verifying signatures on checks and other documents and revealing all types of signature fraud, including random and skilled forgery. Working in all steps of the recognition process: location, noise elimination, segmentation, extraction of the significant features for recognition, data analysis, training of neural networks, merging the results. – Location of machine print, handprint, and cursive courtesy / legal amount in Money Orders and Travel Checks: noise elimination, segmentation, extraction of dollar and cent parts; – Signature verification in scanned images; the results obtained are better than those of human expertise; – Online signature verification; – Location of machine print, handprint and cursive address in scanned images of arbitrarily oriented flats. Работа на всех этапах процесса распознавания: нахождение объекта, очистка от шума, разработка значимых признаков, машинное обучение, оценка результатов. Участие в следующих проектах по распознаванию образов и машинному зрению: Автоматическое распознавание рака груди на маммографических снимках и томограммах (томосинтез) (от поиска до оценки злокачественности) Автоматический поиск адресных блоков на конвертах различных почтовых отправлений для последующего распознавания и сортировки почты; Автоматический детектор поддельных подписей на финансовых документах; Поиск печатных и письменных строк "сумма прописью" на финансовых документах при их распознавании

Taking part in the implementation of a system that blends and represents different types of human motion involving the motion capture technology – Designed and implemented several velocity based blending algorithms. Участие в создании технологии "motion capture", разработка соответствующих алгоритмов.

Participation in the design and implementation of an intelligent Decision Support System for Telecommunication Network Management (by contract with Russia Telecom) (Visual C++6.0, Oracle SQL+ 8.0): – Implemented the detailed GUI design; – Implemented the object-oriented design; – Developed mathematical models and algorithms for the evaluation of telecommunication network efficiency; – As the leader of a team of programmers, was responsible for the modeling, simulation, and GUI implementation of the system.

Implemented the computer visualization of complicated data structures using novel graph drawing algorithms, developed mathematical models (algebraic graph theory, etc.) and implemented programs (Visual C++5.0, SQL) for: – A graph editor toolkit with GUI functions (represented at the International Conference on Social Networks, Vancouver, Canada, 2000); – Software for revealing cyclic references on data structures. Designed and implemented fast algorithms for maintaining and restoring data integrity during data modifications (Visual C++5.0, SQL).

System analysis, developing mathematical models and novel algorithms, implementing programs (C, Visual C++5.0) for the evaluation of the accessibility of graph/network nodes. The results have been applied for the – Evaluation of the accessibility of nodes in transport networks (for a Geo-information system); – Evaluation of centrality/social status and connection/interaction in social networks (for a Sociometric computer system); – Aggregation of incomplete expert preferences (for a Group Decision Support system within an international INTAS project); – In my Ph.D. dissertation. Обучение в аспирантуре, защита кандидатской диссертации.

Participated in the development and implementation of an intelligent Geo-information system for the evaluation of land cost (DBase, Clipper, UI-Programmer, Object Pascal): – Designed and implemented relational databases of city objects; – Developed graphic user interface; – Designed mathematical models for the assignment of centrality attribute to all city objects and setting the land rent based on the relative positions on the geographical, transport, and infrastructure networks. Implemented object-oriented software for the assignment of land rent.