Phd in INFORMATION AND COMMUNICATION TECHNOLOGY (ICT)

Thesis title: Deep Neural Networks for Emotion Recognition and Behavioral Analysis

In the domain of academic research, this doctoral thesis embarks on a rigorous investigation into the utilization of deep neural networks for the dual purposes of behavioral analysis and emotion recognition. At its core, this study grapples with the convergence of advanced technology and the intricacies of human expression. A pivotal focus of the research is the nuanced exploration of human emotions. It represents a critical endeavor as emotions are a foundational aspect of human cognition. The central objective is the deployment of facial landmarks as the primary source of data for emotion classification. This classification hinges on the integration of Convolutional Neural Networks and Long Short-Term Memory networks, harnessing their capabilities for the analysis of emotion within the context of video data. The approach transcends mere classification, striving for a profound interpretation of deep learning models for human emotions studying the inner mechanism behind the emotional feeling. Each facial gesture and micro-expression is meticulously dissected and decoded by these innovative models, effectively converting visual data into quantifiable emotional states. The aim is to decipher the intricate nuances of human emotional expression, mapping these to specific labels within the emotional spectrum. In detail, this project casts a discerning eye toward the aspect of model explainability. In an era where artificial intelligence is heavily relied upon, it is imperative to comprehend the inner workings of these models. The research seeks to elucidate the black box by providing insights into the decision-making processes of these deep neural networks. By rendering the models' decision pathways more transparent, it facilitates a better understanding of their functionality, which in turn has implications for their real-world applications. In the domain of behavioral analysis, the research advances the state of the art. One of the main focuses is on human activity recognition, a pivotal problem in various domains including healthcare, surveillance, and human-computer interaction. A suite of innovative deep learning models emerges, each with the capacity to recognize and classify human activities. These models represent a significant step forward in the field of behavioral analysis, as they demonstrate the potential to classify human actions with a reliable level of accuracy. Moreover, always in this field, numerous clinical applications are presented where deep learning models and artificial intelligence can impact the difficult world of neuro-developmental disorders, being of great help to doctors, psychiatrists, and psychologists for the detection of unfortunate conditions such as autism spectrum disorder, attention deficit and hyperactivity disorder and several other related and common conditions. In summation, this doctoral thesis stands as an exemplar of scientific inquiry. It deftly navigates the intricate landscape of deep learning models, employing them as tools for deciphering the enigmas of human emotion and behavior. Through the systematic examination of facial landmarks, the models capture and classify emotions, offering a deeper understanding of human sentiment. In parallel, the thesis shines a spotlight on model transparency, providing a blueprint for explainable AI. In the realm of behavioral analysis, it forges new frontiers with innovative models, raising the bar for the recognition and prediction of human activities and for the detection of clinical conditions. In essence, this thesis serves to analyze the symbiotic relationship between advanced technology and human comprehension, revealing the inner workings of our emotions and actions through the lens of data-driven analysis.