Paola's research focuses on developing systems capable of compositional reasoning and common-sense inference, using vision and language, while addressing problems regarding cultural biases, data distribution, explainability, and trustworthy AI. Her work also focuses on developing simulated environments to provide a safe, controlled setting where embodied agents can learn. Her aim is to create virtual environments and synthetic agents that can effectively collaborate and solve complex tasks by leveraging the implicit knowledge of large-scale pre-trained deep learning models.