Answering complex queries over incomplete knowledge graphs (KGs) is a challenging job. Most previous works have focused on learning entity/relation embeddings and simulating first-order logic operators with various neural networks. However, they are bottlenecked by the inability to share world knowledge to improve logical reasoning, thus resulting in suboptimal performance. In this paper, we propose a complex reasoning schema over KG upon large language models (LLMs), containing a curriculum-based logical-aware instruction tuning framework, named LACT. Specifically, we augment the arbitrary first-order logical queries via binary tree decomposition, to stimulate the reasoning capability of LLMs. To address the difficulty gap among different types of complex queries, we design a simple and flexible logic-aware curriculum learning framework. Experiments across widely used datasets demonstrate that LACT has substantial improvements~(brings an average +5.5% MRR score) over advanced methods, achieving the new state-of-the-art.

In this paper, we propose a Cross-AttentionTransformer Based Method for Traffic Engineering, CATE, which employs the cross attention mechanism to analyze traffic patterns and dynamically apply different strategies based on traffic characteristics. CATE effectively balances normal network performance with burst traffic resilience, ensuring the delivery of high-quality solutions across various network topologies. Preliminary evaluations on real-world wide-area network (WAN) topology datasets demonstrate that CATE outperforms existing state-of-the-art machine learning-based TE approaches as well as traditional linear programming methods, achieving a significant reduction in average maximum link utilization and improved computational efficiency.

In Industrial Internet of Things (IIoT), timely detection of equipment failures and predictive maintenance are crucial. Leveraging Federated Learning (FL) allows for distributed model training on inspection devices, enabling predictive maintenance without compromising data privacy. However, Traditional FL faces communication and scalability challenges in large scale industrial scenarios. While hierarchical federated learning (HFL) improves flexibility, it struggles in signal-unstable scenarios. This paper proposes a UAV-assisted HFL framework for distributed remote inspection in IIoT, where UAVs enhance communication via high-altitude links and act as edge servers to collect and aggregate model parameters, reducing the central server’s communication burden and improving training efficiency. In this framework, energy-constrained edge clients face challenges of energy efficiency and data silos, while UAV deployment and energy limitations must also be addressed. To optimize fair and energy-saving training, we formulate an optimization problem to minimize energy consumption based on communication and training costs. This is decomposed into two sub-problems: (1) client selection, tackled as a multi-objective optimization using a MAB-based algorithm with a customized reward function balancing energy use and fairness; (2) UAV scheduling, addressed with a heuristic algorithm to optimize edge server deployment. Combining these schemes enables efficient scheduling for large-scale IIoT inspections. Finally, simulation experiments demonstrate the proposed strategy’s significant advantages in reducing system energy consumption, enhancing model accuracy, and improving fairness.