Text-to-SQL, as a hotspot in NLP, is a task of converting natural language text into the structured query language SQL, which makes database interaction more intuitive. This paper presents TC-SQL, a novel framework for schema linking optimization enabled by graph database modeling, with a decoupled table completion modules as its core component. Leveraging the open-source LLM Qwen for table retrieval and a supervised fine-tuned CodeLlama for SQL generation, TC-SQL models database tables and columns within a graph database to enhance recall completeness. Experiments show that it achieves an execution accuracy of 79.3% on the Spider test set, outperforming multiple comparative methods. Ablation studies further confirm the effectiveness of the supervised fine-tuning process and the table recall and completion modules. Leveraging open-source models, the method optimizes schema linking while leveraging open-source models to enhance system controllability and deployability, offering an efficient and reliable solution for intelligent data access in industrial scenarios.

Transformer-based methods for monocular 3D human pose estimation (3DHPE) excel at modeling temporal dependencies. However, existing approaches typically employ a single temporal attention model that treats all body joints as an inseparable whole. This homogeneous treatment fails to capture the distinct motion characteristics of different body parts, such as the trunk versus the limbs, and is inconsistent with the heterogeneous nature of human movement. To address this limitation, we propose the Part-Aware Spatio-Temporal Transformer (PAST-Former). Within a standard spatio-temporal Transformer framework, our model introduces a parallel dualstream temporal attention module. This module decouples motion modeling into two paths: a global path captures the smooth trajectory of the entire body, while a local path performs differential dynamic modeling on semantically divided body parts. The features from these two streams are dynamically fused via an adaptive gating mechanism, allowing the model to flexibly balance global and local information based on the action context. Extensive experiments on the Human3.6M and MPI-INF-3DHP benchmark datasets validate the effectiveness and state-of-the-art performance of our proposed method.

In order to avoid broadcast storms in wireless sensor networks and improve the efficiency of information transmission, virtual backbone networks have been widely introduced into network information transmission. In order to combine the practical development needs, this paper devotes to the optimization research on the construction of virtual backbone network, optimizes the three aspects of routing path length, backbone size and fault tolerance respectively, introduces the GOC-CDS construction strategy and the$\alpha \text{MOC-SCDAS}$construction strategy, and designs an approximation algorithm for the construction of the virtual backbone network based on the two strategies. Through theoretical analysis, the algorithm can obtain a$(72\rho^{2} + 24\rho + m + 1)(2\rho + 1)^{2}$-approximate optimal solution, where$\rho = \frac{r_{max}}{r_{min}}$. After comparative experiments, it is verified that the can maintain a smaller routing path length of the network and a better fault tolerance of the virtual backbone network, and at the same time, it can ensure that the size of the backbone network is as small as possible.

Participatory sensing is often used in environmental or personal data monitoring, wherein a number of participants collect data using their mobile intelligent devices for earning the incentives. However, a lot of additional information is submitted along with the data, such as the participant’s location, IP and incentives. This multimodal information implicitly links to the participant’s identity and exposes the participant’s privacy. In order to solve the issue of these multimodal information associating with participants’ identities, this paper proposes a protocol to ensure anonymous data reporting while providing a dynamic incentive mechanism simultaneously. The proposed protocol first establishes a submission schedule by anonymously selecting a slot in a vector by each member where every member and system entities are oblivious of other members’ slots and then uses this schedule to submit the all members’ data in an encoded vector through bulk transfer and multiplayer dining cryptographers networks (DC-nets) . Hence, the link between the data and the member’s identity is broken. The incentive mechanism uses blind signature to anonymously mark the price and complete the micropayments transfer. Finally, the theoretical analysis of the protocol proves the anonymity, integrity, and efficiency of this protocol. We implemented and tested the protocol on Android phones. The experiment results show that the protocol is efficient for low latency tolerable applications, which is the cases with most participatory sensing applications, and they also show the advantage of our optimization over similar anonymous data reporting protocols.

This paper proposed an optimized strategy of the incentive mechanism for participatory sensing network on the basis of Sense Util which is our previous research work. For optimizing, the strategy considered the possibility of that some users actually cannot participate the current sensing task, so in order to avoid unnecessary energy and bandwidth consumption on server end, users were classified into different subsets according to their position, and server distributed the sensing tasks directionally. To evaluate the performance of the strategy, we conducted simulation study, and the results showed that the optimization work is energy efficient.