Answers are grounded in this paper and cite their sources.
Experimental study demonstrates resilient cross-modal alignment across industrial multimodal datasets, indicating enhanced operational stability for smart manufacturing edge environments.
The rapid development of Industrial Internet of Things (IIoT) systems in Industry 5.0 has led to the continuous generation of massive heterogeneous multimodal data in dynamic manufacturing environments. In such dynamic complex systems, reliable crossmodal alignment is challenged not only by modality gaps, but also by temporal semantic drift, noisy observations, ambiguous feedback, and evolving operating conditions during disturbance, adjustment, and recovery processes. Most existing alignment methods rely on static embedding mappings and therefore cannot adequately capture long-term semantic evolution or handle uncertainty in adaptive industrial scenarios. To address these limitations, this study proposes Semantic Trajectory-based Collaborative Learning (STCL), a lightweight and robust cross-modal learning framework for resilient IIoT systems. The primary objective of STCL is cross-modal alignment under dynamic temporal evolution, while temporal modeling and human feedback are introduced to improve semantic consistency and robustness. STCL combines semantic trajectory alignment with a human-guided consistency learning strategy to support feedback-driven correction of semantic drift enhanced by a hybrid fuzzy-Bayesian perspective. The trajectory module models the temporal evolution of multimodal representations to preserve long-term semantic consistency, while the guidance module incorporates sparse human feedback as uncertainty-aware supervisory signals to mitigate semantic drift, improve interpretability, and support adaptive semantic correction. In this way, STCL jointly captures local alignment quality, global temporal consistency, and human-centered semantic adjustment within a unified optimization framework. Experiments on four public multimodal datasets demonstrate that STCL consistently outperforms representative baseline methods in alignment robustness, accuracy, and temporal stability. In addition, STCL maintains low computational overhead, indicating its suitability for resource-constrained IIoT edge environments. Overall, STCL provides an efficient, human-centric, and uncertaintyaware solution for resilient cross-modal learning in Industry 5.0-enabled IIoT systems, with potential value for intelligent perception, collaborative response, and simulationoriented industrial decision support.
No takes yet. Share an insight, caveat, or question.
Sang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: