Subject Ph. D.

The integration of symbolic and connectionist approaches to artificial intelligence represents a significant frontier in achieving explainable cognitive models in Human-Ai systems and in particular in deep learning systems. The rapid advancements in deep learning have led to significant breakthroughs in various fields, including natural language processing, computer vision, and autonomous systems. However, a major limitation of these models is their lack of interpretability. Unlike human cognition, which relies on structured, symbolic reasoning and explicit knowledge representation, deep learning systems function as "black boxes," making their decision-making processes difficult to understand and verify. This lack of transparency is particularly problematic in high-stakes applications such as medicine, law, finance, and industrial automation and cyber physical social systems, where explainability is critical for ensuring trust, accountability, and regulatory compliance. This PhD project aims to bridge the gap between symbolic and connectionist AI by developing a hybrid AI framework that integrates formal logics, ontologies, and structured knowledge representations into deep learning architectures. By combining the power of symbolic reasoning with the learning capabilities of deep neural networks, the proposed research seeks to enhance AI's interpretability while maintaining high predictive accuracy. The objective is to create AI systems capable of explicit reasoning and human-comprehensible decision-making, enabling their adoption in fields where trust and explainability are paramount. The research will focus on four key areas. First, it will develop formal cognitive knowledge representations, leveraging Formal Concept Analysis, ontologies, and logical frameworks to structure domain knowledge and enforce reasoning constraints. Second, it will explore hybrid learning strategies that integrate symbolic constraints into deep learning, allowing neural networks to incorporate prior knowledge while refining their representations through data-driven learning. Third, the project will investigate explainability mechanisms, including knowledge graphs, rule-based reasoning, and attention mechanisms, to provide transparent and human-understandable justifications for AI-generated decisions. Finally, the PhD will focus on validating the framework in real-world applications ensuring its robustness, scalability, and compliance with domain-specific interpretability requirements. To achieve these goals, the project will employ a multi-disciplinary approach, integrating research in symbolic AI, machine learning, cognitive science, and knowledge engineering. It will develop neurosymbolic architectures capable of performing structured reasoning, ensuring that AI-generated conclusions align with human cognitive processes and expert knowledge. Furthermore, the research will explore differentiable symbolic reasoning frameworks, allowing deep learning models to incorporate logical rules and hierarchical knowledge structures seamlessly. The expected contributions of this PhD include the development of a novel hybrid AI framework for explainable cognition, the introduction of new methodologies for integrating symbolic AI with connectionist models, and the definition of novel metrics for measuring interpretability, verifiability, and robustness in AI systems. The research will also present case studies demonstrating the framework's effectiveness in domains where explainability is important such as the (AI)-driven framework to dynamically integrate heterogeneous real data sources, as industrial Cyber Physical Systems and formalize the semantics clusters for human-centric intelligent enterprise systems of the MG-IB enterprise that participates in the Human-AI chair. The symbolic knowledge will be based on rules of physics such as those found in industry. By harmonizing the strengths of symbolic AI and deep learning, this project aspires to establish a new standard for explainable artificial intelligence, paving the way for AI models that are not only powerful and data-driven but also transparent, interpretable, and aligned with human reasoning expectations.