Guangdong Baiyun University

Despite China's strategic emphasis on AI education, gaps remain in evaluating undergraduates' competencies and applying international frameworks in this context. The UNESCO AI Competency Framework for Students lacks empirical validation, especially in non-Western settings. This study validates it using survey data from 583 undergraduates across 13 institutions in 9 Chinese cities. We examined four dimensions: Human-centered mindset, Ethics of AI, AI techniques and applications, and AI systems design. Students showed stronger competencies in mindset and ethics than in technical areas. First-years reported higher technical skills than seniors, challenging progression models. Principal component analysis confirmed the four-dimensional structure, with multivariate analyses revealing limited gender and discipline effects, mainly in technical aspects. Findings highlight a theory-practice gap misaligned with national priorities for practical skills. We recommend curriculum reforms via AI co-creation, spiral design, and personalized pathways for responsible AI engagement. This advances theoretical insights and practical guidance for higher education.

Recent studies have challenged the notion that bone tissue lacks lymphatic vessels, highlighting their essential role in bone regeneration. Effective bone repair relies on the interplay of osteogenesis, angiogenesis, and lymphangiogenesis. Here, we present a three dimensional-printed biomimetic Gelatin methacryloyl/icariin@Mesoporous silica nanoparticle/Hydroxyapatite (GelMA/ICA@MSN/HAp) scaffold designed to recruit bone marrow mesenchymal stem cells (BMSCs), human umbilical vein endothelial cells (HUVECs), and lymphatic endothelial cells (LECs), enabling these cells to self-assemble into "cell islands" that coordinate tissue regeneration. The scaffold exhibits excellent biocompatibility, biodegradability, and sustained cytokine release. In vitro, it promotes the migration, proliferation, and lineage-specific differentiation of BMSCs, HUVECs, and LECs. In an osteoporotic bone defect model, the scaffold significantly enhances new bone formation, supports vascular remodeling by recruiting HUVECs, and induces lymphatic maturation via LECs, accompanied by upregulation of bone morphogenetic protein 2, vascular endothelial growth factor, and prospero homeobox protein 1. Transcriptomic analysis identifies activation of the hypoxia-inducible factor 1 alpha (HIF-1α) signaling pathway as crucial to these effects. Mechanistically, LECs-derived conditioned medium stimulates HUVECs angiogenesis and BMSCs osteogenesis by inducing HIF-1α mediated mitochondrial metabolic reprogramming. This is the first study to integrate lymphatic modulation into scaffold design for bone repair, achieving osteo-angio-lymphogenic coupling and establishing a novel organoid model for bone regeneration.

Quantized deep neural networks (QDNNs) provide an efficient solution for real-time human activity recognition (HAR) on resource-constrained edge devices by addressing computational and memory limitations while maintaining high classification accuracy. This study proposes a QDNN-based framework utilizing advanced quantization techniques, including weight pruning, fixed-point quantization, and layer compression, to optimize model size, inference speed, and resource efficiency. Additionally, the integration of optical sensors for high-resolution data acquisition further enhances the robustness and sensitivity of activity recognition systems. An adaptive quantization-aware training strategy is integrated into the framework to mitigate accuracy loss caused by quantization and ensure robust performance in dynamic environments. The framework was evaluated on benchmark datasets, UCI HAR and WISDM, using Raspberry Pi and NVIDIA Jetson Nano as deployment platforms. Experimental results demonstrate a 45% reduction in memory consumption, a 38% decrease in inference latency, and up to a 42% reduction in computational complexity compared to baseline DNNs, while maintaining over 98% classification accuracy across various quantization levels. These results highlight the practicality of QDNNs for enabling real-time HAR in wearable health monitoring, smart homes, and other edge-based applications, offering a scalable and efficient approach to deploying machine learning models in resource-limited settings.