Диссертации, представленные на защиту и подготовленные в НИУ ВШЭ

Диссертация посвящена трём взаимосвязанным компонентам самоконтролируемого обучения: архитектурам моделей, функциям потерь и специфичным для данных вспомогательным задачам, применительно к визуальным данным, временным рядам с промышленных датчиков и графовым данным. Для визуальных данных предложены архитектуры на основе рекуррентных сетей с механизмами внимания и визуальных трансформеров для монокулярной оценки глубины, включая облегчённую модель для мобильных устройств. Отдельно разработаны подходы к визуальному распознаванию места на основе семантической и глубинной информации и фреймворк дистилляции знаний «учитель-ученик» для обнаружения 3D-объектов. В части временных рядов с промышленных датчиков разработаны специализированные функции потерь и архитектуры моделей взаимодействия, которые позволяют устойчиво диагностировать неисправности и предсказывать остаточный ресурс промышленного оборудования при минимальном объёме размеченных данных. На графовых данных предложены новые вспомогательные задачи: построение векторных представлений рёбер статических и динамических графов, обучение на молекулярных графах и формирование вознаграждения в обучении с подкреплением на основе графовых нейронных сетей. Результаты расширяют теоретическую базу самоконтролируемого обучения и апробированы в задачах промышленных партнёров. В их числе интеграция в задачах индустриальных партнеров.

This dissertation examines why oil-producing economies with comparable hydrocarbon endowments and access to similar technological opportunities follow divergent innovation trajectories. It develops the Knowledge Development Ecosystem Model (KDEM), which treats institutions, culture, and technological learning as configurationally interacting rather than additively independent dimensions, and identifies knowledge codification by the system's dominant integrator; oilfield-services firms in market and hybrid systems, national oil companies in centralized state-led systems, as the previously unspecified channel linking these dimensions to innovation outcomes. The study combines comparative case analysis of Norway, the United States, Saudi Arabia, and Russia (2004–2021), two-way fixed-effects panel estimation, and firm-level Wright-type learning curves for Bakken shale operators. Institutional quality is associated with petroleum-sector innovation primarily through a substitutive interaction with technological capability: where technological capability is low, institutions are the binding margin, and where it is high, further institutional gains add little. This pattern is sign-stable under drop-one-country re-estimation and anchored by the Saudi Arabia–United States contrast; cultural moderation is not separately identified at this sample size. The findings highlight the configurational interaction of institutional and technological conditions and the role of oilfield-services firms and national oil companies as knowledge integrators in resource-based innovation systems.

Given the importance of innovation for organizational competitive advantage, developing innovative abilities among the current and future workforce has become a pressing human resource priority. This dissertation investigates how organizational and personal factors shape innovative behaviour using multiple samples—employees in Nigeria across 13 industries and higher education students in Russia. Using Latent Profile Analysis (LPA), the study identifies three distinct profiles of student innovators, while multinomial logistic regression (MLR) results show that individual resilience and self-monitoring are significant positive predictors of membership in the more innovative profiles. Further analyses using Structural Equation Modelling (SEM) on the employee data generate three key insights. First, employees whose supervisors hide knowledge tend to exhibit lower innovative work behaviour because they imitate such behaviour and experience weakened working relationships with colleagues. Second, morally disengaged employees engage less in innovative behaviour because they are more likely to hide knowledge and damage workplace relationships. Finally, employees who perceive their jobs as offering flexible working opportunities display higher innovative work behaviour because they communicate more frequently with their supervisors and engage in less knowledge hiding. Collectively, these findings provide important implications for supervisor leadership behaviour, the design of flexible work arrangements, and curriculum development aimed at fostering innovative behaviour in both organizational and educational contexts.

The mirror neuron system (MNS), first identified in primates and later confirmed in humans, activates during both action execution and observation, linking perception and motor control. While most studies emphasize excitatory facilitation, inhibitory mechanisms remain understudied. This dissertation investigates the dual excitatory–inhibitory nature of the MNS, its temporal dynamics, and contextual robustness, using transcranial magnetic stimulation (TMS) across four experimental studies.A novel Postvideo paradigm, involving TMS after observed action completion, elicited the most stable mirror responses compared to traditional static or video conditions. Results demonstrated a dual mechanism: facilitation in task-relevant muscles and suppression in non-related muscles, extending motor surround inhibition (mSI) from action execution to observation. Temporal analyses revealed that inhibitory effects dominate at later stages (up to 640 ms and post-movement), highlighting the critical role of post-movement phases. Cortical excitatory and inhibitory circuits (ICF, SICI) were differentially modulated, suggesting GABAergic contributions. Importantly, mirror responses proved robust across variations in body posture, confirming independence from proprioceptive state.The novelty of this work lies in reframing the MNS as a balanced excitatory–inhibitory network, demonstrating the centrality of inhibition to mirror responses, and introducing the Postvideo paradigm as a methodological advancement. The findings deepen theoretical understanding of mirror response, while offering practical implications for neurorehabilitation of motor-impaired patients.

The thesis is devoted to the study of features of Anti-Western currents of thought in the context of the decline of Western hegemony and crisis of the liberal international order. The thesis provides a comparative analysis of Neo-Ottomanism and Neo-Eurasianism in order to reveal the patterns that are specific for Anti-Westernism in the beginning of the 21st century. As a response to the quest for new identities and international position of Turkey and Russia in the new world order, Neo-Ottomanism and Neo-Eurasianism appeared following the end of the Cold War and became more prominent since 2000. Their rise occurred against the backdrop of the shifts in Turkish and Russian foreign policies under the personalized rules of Erdoğan and Putin. Neo-Ottomanism and Neo-Eurasianism offered macro-political identities, assertive foreign policy visions and geopolitical concepts as an alternative to the Western ones. Underlining the importance of ideas in world politics, the thesis employs constructivist approach and the concept of worldview. The comparative analysis of Neo-Ottomanist and Neo-Eurasianist worldviews demonstrated that the patterns of contemporary Anti-Westernism encompass opposition to the Western dominated world order, criticism of Western philosophy, thought and conservative response to Western modernity, secularism and liberal values. These two Anti-Western and conservative worldviews offer an alternative to the West based on civilizational and conservative values including specific moral principles, emphasize on the religions, believe in strong authority and the imagine of supranational identity.

This research focuses on developing methods for converting raster images and three-dimensional objects into vector representations using deep learning. Vectorization of objects involves finding object representations using mathematical primitives and relationships between them.To achieve this goal, the following tasks were addressed: data collection, construction of mathematical models, and development of vectorization algorithms. Data collection was performed by processing scanned images of 2D and 3D objects and generating synthetic data. Automatic algorithms using computer vision methods were developed for data cleaning and processing, along with procedures for manual data processing. These algorithms facilitate semi-automatic annotation of data, opening up the possibility to train neural networks using deep learning methods. Various neural network architectures, including convolutional neural networks and transformers, are explored to create models capable of accurately and efficiently vectorizing technical drawings and 3D point clouds. The proposed algorithms demonstrate high accuracy and efficiency in solving object vectorization tasks, with potential applications in computer vision, robotics, and data visualization.

The dissertation is comprised of three separate papers on one overarching topic -- causes and consequenses of economic inequality. I investigate inequality both at the cross-country and within country level. Specifically, I explore the possible channels, such as the financial sector, through which within-country income inequality could affect macroeconomic outcomes. I also explore such causes of cross-country income differences as reserve currency status, and colonial history.

The purpose of this research is to develop econometric models in order to reveal observable and unobservable heterogeneity in economic choices by producers and consumers, to disentangle heterogeneous effects of exogenous shocks on firm costs, and to evaluate heterogeneous effects of policy reforms aimed at price and quality regulation. The theoretical part of the dissertation deals with conditional quantile regression, data envelopment analysis and dynamic panel data models. The empirical part of this dissertation employs conditional quantile regression, finite mixture (latent class) models, stochastic frontier analysis, data envelopment analysis, dynamic panel data models and treatment effect estimation in an application to several economic fields: banking, productivity analysis, health economics, contracts and empirical corporate finance. As regards regulation, the models help to identify the heterogeneous effect of policy reforms on different agents. Extensions of the methodology pertain to empirical labor economics, energy economics, municipal finance and macroeconomics.

Nowadays, recommender systems are essential components of various consumer services, from e-commerce to social media. They help navigate through a large volume of items empowering user experience. Methods vary from classic matrix completion techniques to modern sequence models inspired by natural language processing. One of the prominent approaches is to consider the recommender system as a link prediction problem on a bipartite user-item interaction graph.The dissertation studies the adaptation of network embedding techniques to recommender systems. In the dissertation, we investigated different aspects of graph machine learning techniques, and their effect on downstream link prediction problem. We proposed an efficient strategy to incorporate node and edge features with structural information to solve the link prediction (recommendation) problem. Also, we developed novel models to preserve temporality and enable graph-based exploration for recommender systems. Proposed approaches were compared with the state-of-the-art open-source benchmarks and showed the efficiency of our approach.

Our aging world is faced with the ever-growing incidence of dementia and age-related cognitive decline. In the absence of satisfactory pharmacological treatments, it becomes increasingly important to identify alternative, lifestyle-based approaches to preventing and mitigating age-related cognitive impairments. The leading theoretical framework in this area is that of so-called cognitive reserve (CR). CR is defined as the level of resilience of an individual’s cognitive processes to age-related brain deterioration. One experience-related factor that has been linked to CR build-up is bilingualism: by means of a constant additional cognitive effort required to control dual-language use, bilinguals strengthen their neurocognitive substrate which provides a protective effect against age-related cognitive decline, mediated by the CR mechanism. In my PhD research, I examined the role of individual life experiences, bilingualism in particular, in promoting the development of CR during the entire lifespan. Moreover, to tackle the long-standing issue of CR measurability, I refined recent approaches to devise a novel method for quantifying CR directly. For the first time in the literature, I tested its validity by comparing it with traditional approaches to measuring CR. This newly-developed measure may contribute to advancing the theoretical framework of CR, as well as inform the clinical practice, enabling new approaches in the fight against cognitive aging through the design of personalized care plans. Individuating CR-enhancing factors and their underlying mechanisms may provide a tool to improve senior citizens’ quality of life, and ease the economic burden on healthcare systems and public funds.