Knowledge-Infused Learning: A Sweet Spot in Neuro-Symbolic AI

Deep learning has revolutionized the artificial intelligence (AI) landscape by enhancing machine capabilities to understand data-dependant relationships. On the other hand, knowledge may not directly correlate or depend on the data but represents facts that are true. Combining knowledge with the data-driven deep learning techniques improves upon what can be learned from data alone, resulting in improved performance with reduced training, user-level explainability, modeling uncertainty in deep learning, achieving context-sensitivity, and better control over the behavior of AI systems such as to assure the safety or avoid toxic behavior. We refer to the approach of combining various types of explicit knowledge as knowledge-infused learning (KiL). Knowledge infusion brings symbolic AI into data-driven AI, giving us a class of neuro-symbolic AI methods. The work on KiL has already developed a suite of context-adaptive algorithms that infuses various knowledge into deep learning methods in various ways, broadly categorized as a shallow infusion, semi-deep infusion, and deep infusion. This special issue allows interdisciplinary researchers and practitioners from diverse fields such as natural language processing, recommender systems, and computer vision to contribute their research on the infusion of external and expert-curated knowledge in data-driven learning methodologies for consistency and robustness in outcomes.