Neurosymbolic Framework for Concept-Driven Logical Reasoning in Skeleton-Based Human Action Recognition

Researchers introduce a neurosymbolic framework that combines neural networks with symbolic logic for skeleton-based human action recognition, enabling interpretable AI models that explain their decisions through human-readable logical rules rather than operating as black boxes.

This research addresses a critical challenge in AI interpretability by bridging the gap between neural networks and symbolic reasoning. Skeleton-based human action recognition typically relies on deep learning models that achieve strong performance but lack transparency in their decision-making processes. The neurosymbolic approach reframes action recognition as concept-driven logical reasoning over motion primitives, grounding abstract mathematical operations in semantically meaningful concepts. The framework employs a spatio-temporal skeleton encoder to extract motion representations, then maps these to interpretable concept predicates through a specialized decoder that separates pose-centric and dynamics-centric abstractions. By anchoring skeleton representations with language model descriptions of atomic motion primitives, the system establishes a shared conceptual space between perception and reasoning layers. This alignment ensures that learned concepts remain semantically coherent and human-understandable. The experimental validation on benchmark datasets (NTU RGB+D 60/120 and NW-UCLA) demonstrates that competitive recognition performance is maintained while providing explicit logical explanations for predictions. This work exemplifies a broader trend in AI toward explainable systems, particularly important for applications in surveillance, healthcare, and robotics where decision transparency matters. The neurosymbolic paradigm offers developers a pathway to deploy action recognition systems that users and regulators can audit and trust, rather than relying solely on accuracy metrics. As AI adoption accelerates across safety-critical domains, frameworks enabling both performance and interpretability will become increasingly valuable for building trustworthy systems.

• →Neurosymbolic framework enables skeleton-based action recognition with human-readable logical explanations instead of black-box predictions.
• →First-order logic layers enable models to learn interpretable rules governing action semantics through differentiable reasoning.
• →LLM-derived descriptions ground motion concepts in semantic space, ensuring learned predicates remain meaningful and auditable.
• →Competitive performance on standard benchmarks validates that interpretability does not require sacrificing recognition accuracy.
• →Approach addresses industry need for transparent AI systems in surveillance and healthcare applications where decision explanations are critical.