[2605.07140] Neurosymbolic Framework for Concept-Driven Logical Reasoning in Skeleton-Based Human Action Recognition

Computer Science > Computer Vision and Pattern Recognition arXiv:2605.07140 (cs) [Submitted on 8 May 2026] Title:Neurosymbolic Framework for Concept-Driven Logical Reasoning in Skeleton-Based Human Action Recognition Authors:Talha Ilyas, Deval Mehta, Zongyuan Ge View a PDF of the paper titled Neurosymbolic Framework for Concept-Driven Logical Reasoning in Skeleton-Based Human Action Recognition, by Talha Ilyas and 1 other authors View PDF HTML (experimental) Abstract:Skeleton-based human activity recognition has achieved strong empirical performance, yet most existing models remain black boxes and difficult to interpret. In this work, we introduce a neurosymbolic formulation of skeleton-based HAR that reframes action recognition as concept-driven first-order logical reasoning over motion primitives. Our framework bridges representation learning and symbolic inference by grounding first-order logic predicates in learnable spatial and temporal motion concepts. Specifically, we employ a standard spatio-temporal skeleton encoder to extract latent motion representations, which are then mapped to interpretable concept predicates via a spatio-temporal concept decoder that explicitly separates pose-centric and dynamics-centric abstractions. These concept predicates are composed through differentiable first-order logic layers, enabling the model to learn human-readable logical rules that govern action semantics. To impose semantic structure on the learned concepts, we align skeleto...