Titulo Estágio
Decoding Perceived Valence from Brain Signals with Graph Neural Networks
Áreas de especialidade
Sistemas Inteligentes
Local do Estágio
CISUC-DEI
Enquadramento
Understanding people's emotions plays a crucial role in daily human communication, mental health, and overall quality of life. Recognizing users' emotions is essential for advanced human-computer interaction in many emerging applications. Effective emotion recognition and regulation are critical human skills for sustainable mental health, performance, and quality of life. Failure in this area can lead to underperformance, disengagement, and debilitating disorders. Mental health disorders encompass a wide range of diagnoses, including developmental disorders like Attention Deficit Hyperactivity Disorder (ADHD), substance use disorders such as alcohol dependence, and depressive and anxiety disorders.
With the advancements in machine learning, particularly in emotion recognition, there is potential for developing novel neurorehabilitative strategies to minimize mental health problems.
However, the opaque black-box nature of most machine learning models poses substantial ethical, legal, and governance challenges. Addressing these concerns is essential for fostering trust and ensuring the ethical use of AI models. Graph Neural Networks (GNNs) have emerged as a powerful technique to leverage both the graph’s connectivity and the input features on various nodes and edges, offering a pathway to bridge the gap between results and reasoning.
Objetivo
The primary goal of this project is to leverage the capabilities of Graph Neural Networks (GNNs) to analyze and predict the valence (emotional value) perceived by the brain for various stimuli. By representing stimuli as graph structures, we aim to develop a robust model that can accurately identify and classify the emotional impact of different inputs in functional Magnetic Resonance Imaging (fMRI) data.
Plano de Trabalhos - Semestre 1
- State-of-the-art review
- Exploration of the fMRI datasets
- Definition of functional networks of interest
- Initial exploration of GNN frameworks
- Writing of the intermediate report.
Plano de Trabalhos - Semestre 2
- Continue exploration of GNN frameworks
- Optimize models
- Explore the explainability and biological correlates of the results
- Prepare a scientific article;
- Prepare the documentation and final Dissertation.
Condições
Access to two in-house datasets of emotion-inducing stimuli (one based on music and another based on videoclips)
Observações
Supervisors:
Marco Simões
Bruno Direito
Daniel Agostinho
Orientador
Marco Simões
msimoes@dei.uc.pt 📩