Automatic Recognition of Psychophysiological States Based on Multimodal Data Analysis
Description: The development of methods for automatic recognition of human psychophysiological states based on multimodal data analysis represents one of the priority directions in contemporary technical science. It finds broad applications in telemedicine systems, human-computer interfaces, educational technologies, and digital mental health services. Computational approaches that enable the detection of various emotional states, as well as stress and depression, based on audio and physiological signals under conditions of limited access to expert diagnostics, high privacy requirements, and multilingual diversity are of particular importance.
At the current stage of the discipline, the primary focus lies on deep learning models, which demonstrate high accuracy when large volumes of annotated data are available. However, such methods have a number of significant limitations: they are characterized by high computational complexity and lack interpretability, which restricts their use in tasks related to psychophysiological assessment, especially in clinical or user-sensitive scenarios.
Thus, the relevance of this research is determined by the necessity to create computationally efficient and interpretable methods for multimodal data analysis that ensure high-quality recognition of psychophysiological states and suitability for practical use in various applied scenarios.
Description: This research focuses on developing and evaluating transformer-based architectures for automatic depression detection using speech data. The study investigates both linguistic and acoustic characteristics of speech and explores their effectiveness in identifying depression across different cultural and linguistic backgrounds. By leveraging advanced deep learning models and cross-cultural datasets, the work aims to improve the accuracy, generalizability, and robustness of speech-based mental health assessment systems.
Description: This thesis investigates the use of Genetic Programming to create explainable machine learning models for depression prediction. The approach focuses on producing symbolic, human-readable models and applying interpretation techniques that clearly communicate the decision-making process to both mental health professionals and the general public. The aim is to combine strong predictive performance with transparency, fostering trust and understanding among all stakeholders.
