Neuro-Symbolic AI for Advanced Signal and Image Processing: A Review of Recent Trends and Future Directions

Abstract

Neuro-Symbolic Artificial Intelligence (NeSyAI) is a paradigm that combines neural networks with symbolic reasoning, building upon foundations laid in the 1990s and early 2000s, yet gaining increased attention through practical applications. This paper provides a comprehensive review of NeSyAI techniques and their applications in advanced signal and image processing. The paper begins by introducing the fundamentals of NeSyAI and a taxonometry analysis, highlighting how it bridges the gap between data-driven learning and knowledge-based reasoning. It then surveys several key application domains, such as biomedical, autonomous robotics, and IoT, in order to understand the benefits and challenges of that integration. For each domain, how NeSyAI methods improve upon traditional purely neural or purely symbolic approaches is illustrated. A comparative analysis with conventional AI techniques is presented, underscoring the advantages of NeSyAI in terms of interpretability, generalization, and flexibility. The challenges that arise in developing NeSyAI systems, such as computational complexity, integration of heterogeneous models, and ethical considerations, are also discussed. Finally, future research trends in NeSyAI for signal and image processing, as well as the path toward more explainable and generalizable AI, are synthesized.