Диссертации, представленные на защиту и подготовленные в НИУ ВШЭ

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.

Исследование посвящено разработке алгоритмических средств анализа электрической активности головного мозга. Предложены новые подходы к оценке распределения и динамики нейрональных источников и функциональных сетей на основе обработки неинвазивных ЭЭГ и МЭГ измерений активности головного мозга. Разработана методология анализа МЭГ данных пациентов с эпилепсией с целью обнаружения эпилептогенной коры на основе анализа пространственной динамики локальной и распределенной нейрональной активности. Особое внимание уделено решению задачи декодирования нейрональной активности и приведены примеры применения разработанных  подходов при создании  моторных и речевых нейроинтерфейсов. В основе предлагаемых решений  - компактная нейронная сеть, для которой получены и верифицированы методики интерпретации весовых коэффициентов в призме существующих в нейронауке знаний о происхождении нейрональной активности и физике процесса ее регистрации. Эмпирически исследовано влияние задержки в предъявлении  сигнала обратной связи на эффективность технологии нейрообратной связи и предложены новые алгоритмы оценки параметров ритмической активности головного мозга в реальном времени, использование которых позволяет сократить задержку в предъявлении сигнала обратной связи и повысить эффективность терапии.

Numerous cognitive studies have demonstrated experience-induced plasticity in the primary sensory cortex, indicating that repeated decisions could modulate sensory processing. In this context, we investigated whether an auditory version of the monetary incentive delay (MID) task could change the neural processing of the incentive cues that code expected monetary outcomes. To study sensory plasticity, we presented the incentive cues as deviants during auditory oddball sessions recorded before and after training during the two MID task sessions, where we focused on the changes in mismatch negativity (MMN) and P3a components. We found that negative outcomes were accompanied by appearance of the feedback-related negativity (FRN), which amplitude was modulated by probability and magnitude of the monetary outcome. In the oddball sessions, we found that MMN and P3a components changed their amplitudes after 2 sessions of the MID task. At the individual level, the training-induced change of mismatch-related negativity was correlated with the amplitude of the feedback-related negativity (FRN) recorded during the MID task. Our results show that the MID task evokes plasticity changes in the auditory system associated with better passive discrimination of incentive cues and with enhanced involuntary attention switching towards these cues.

Systems that process electrophysiological measurements of brain activity in real time, in particular, brain-computer interfaces (BCIs), are used in a number of fields for various research and practical purposes. Most often, such systems use non-invasive recording methods, which are safe and available but introduce limitations in signal quality and throughput that bound the complexity of BCIs that can be implemented with this technology. Invasive recording techniques, on the other hand, give access to more informative signal but pose their own risks and limitations related to the required medical procedures. A currently emerging trend in BCI development is using electrocorticogram (ECoG), an invasive method that is considered to be safer than microelectrode implantation and is routinely used in clinical practice. ECoG yields high spatial resolution, low noise levels, near absence of oculographic and myographic artifacts, proximity to the signal sources and possibility to conduct cortical stimulation through the electrodes, all of which are desirable characteristics for implementation of a complex BCI. In our work, firstly, we propose methods utilizing additional information to improve decoding in non-invasive interfaces, and, secondly, develop experimental paradigms and signal processing methods for ECoG, allowing to perform real-time finger movement decoding, passive cortical mapping, and explore the artificial haptic feedback that could be carried out through ECoG stimulation.