Hybrid storage has been widely used to combine fast and slow storage for a better trade-off between cost-effectiveness and high performance. Integrating the emerging zoned storage devices into hybrid storage and exposing the user-friendly interface to the upper-layer applications can provide better cost-effectiveness and control capabilities. However, legacy hybrid storage approaches are used for block-based devices. Adopting legacy hybrid storage approaches has several challenges, including limited flexibility in zone management, inefficient garbage collection, and penalties of performance and endurance. To overcome those challenges, we propose HyzoneStore, a zone-based hybrid storage with a flexible logical interface and optimized cache. In HyzoneStore, we propose using fast Zoned Namespace SSD (ZNS SSD) as the zoned caching tier for Shingled Magnetic Recording drives (SMR drives) to improve performance while reducing total costs. Our evaluation shows that HyzoneStore can achieve up to 75% throughput improvement compared with SMR drive-based storage. At the same time, we effectively reduce up to 50% of the write amplification on ZNS SSDs and improve its endurance.
In order to solve the problem of precise matching between students and posts in the employment of universities, a employment intelligence service platform is designed to meet the needs of universities, students, and employers. This platform is based on advanced technologies such as big data and artificial intelligence, and collects, integrates, analyzes, and mines employment related data. By constructing a student employment portrait and a two-way employment recommendation model, it achieves precise matching between student needs and employers, accurate promotion of job information, and creates a full chain, one-stop, and humanized precision employment service big data platform, further improving the level of smart employment services in universities.
When measuring and inspecting hull surfaces using a 3D laser scanner, the large volume and complex surface features of the hull necessitate multi-station scanning. Complete hull surface data is obtained by registering point cloud data from multiple stations. However, the registration process accumulates errors, resulting in the registration not meeting inspection accuracy requirements. To address the issue of cumulative errors in point cloud registration for large-scale laser scanning, a multi-station point cloud data registration error correction algorithm based on global control targets is proposed. This algorithm leverages the long-distance measurement and high accuracy of total stations to compensate for the large measurement errors of the scanner at long distances. By applying the coordinate information of global control targets measured by the total station to the point cloud data registration, cumulative registration errors are effectively avoided, thereby improving measurement accuracy. Experimental results show that the proposed algorithm reduces the cumulative error of the first and last point clouds to 16% of that of traditional registration, effectively eliminating registration errors and meeting inspection requirements.
Along with the development of ICT technology, physical and building security services are also evolved. State-of-the-art sensors detect the slightest changes in movement, heat, smoke, light, vibration, etc. Most of the on-premise servers were used to provide building care services in the existing village or small city-scale areas. In this environment, we were able to provide reliable services to customers without any problems. However, there is a problem in providing services for large cities or the entire country. There is a limit to storing and analyzing a big amount of alarm data in the on-premise server. Furthermore, the detection mechanism determines actual intrusion based on pre-defined rule without data analytics. This mechanism has high false alarm ratio and error rate. To solve this problem, this paper introduces a public cloud-based platform that provides security services by collecting, storing, and analyzing alarm data simultaneously generated across the entire country. The proposed platform improves the accuracy of false intrusion determination through machine learning in public cloud. Compared to previous rule-base algorithm, our model improves false alarm detection ratio around 20%. It uses both structured and unstructured data sets to determine false alarms. This platform supports the security guard (commander) by visualizing the analysis results using Microsoft Power BI service. In addition, we provide statistical analysis results between alarm data and related weather conditions.
Online reviews are a valuable source for understanding tourist satisfaction and their emotional tendencies towards attractions. However, there is a need to improve the quality screening of reviews before conducting sentiment analysis. This paper focuses on the important attractions in Macao and utilizes data collected from Ctrip.com to establish an evaluation system. The system scores and ranks the validity of online reviews, and various machine learning methods are examined to automate the screening process. The study finds that XGBoost + stacking is the most effective method, with the upper quartile serving as the threshold for review selection. By employing text sentiment analysis technology, the study evaluates visitor satisfaction for each attraction using quality reviews. This approach contributes to the development of a more comprehensive satisfaction evaluation system and offers a fresh analytical perspective for studying Macau tourism.
With the advent of the big data era, the appraisal system of news discourse is also facing significant changes. Especially in Chinese cultural news report, the application of big data has brought new perspectives and dimensions to the appraisal system. This study will apply big data technology to explore the resources of Chinese cultural news discourse, and deeply analyze the application of big data in the engagement system, attitude system, and graduation system, in order to provide references for related research. It hopes that big data application provides new ideas and methods for the further researches of appraisal system of Chinese cultural news discourse.
Review and comb through the literature in the field of Smart Elderly Care and then make related visualized analysis to explore the research hotspots and development trends in the field of Smart Elderly Care in China and provide references for the development of the field in China. With the employment of the literature in the field of Smart Elderly Care published by CNKI in 2018-2023 as a search sample, the keyword fields of the literature were cleaned by Bicomb; Meanwhile, the keywords were bi-directionally clustered with the help of the visualization software gCLUTO, and the knowledge graph was also presented by using CiteSpace to detect the leading research fronts in the field. After screening, 1,986 articles of related literature were obtained, and 6 clusters were analyzed by cluster analysis, which were classified into 3 research themes, namely, “Smart Elderly Care”, “Pension Institutions”, and “Home-Based Elderly Care”. With the implementation of the “14th Five-Year Plan”, China's Smart Elderly Care has developed rapidly and been getting matured, combined with a clearer research focus and diversified research fronts.
Clothing image search serves as a pivotal technique for efficiently retrieving the most relevant clothing items based on customer queries. The prevailing approaches consist of two serial steps, which first detect and crop out the target clothing, and then feed into a network for similarity learning. Alternatively, there are also one-step approaches that incorporate clothing detection and search within an end-to-end framework. But this approach may encounter optimization contradictions during the training phase, given the disparate optimization requirements of the clothing detection and Re-ID branches. In this work, we propose an end-to-end network for joint clothing detection and search by adding a Re-ID branch based on the attention mechanism in parallel. At the same time, in order to alleviate the conflict between the two branches, we adopt a global context decoupling module to effectively separate the deep features obtained by the backbone network into different branch-specific representations. Extensive results verify the superiority of our proposed method.
Sheet rolling finishing is the most commonly used machining method for single curvature plate. Conventional manual sample testing method cannot realize the automation and digitization of forming processing. The discrete point data of sheet metal forming is obtained by laser detection device. However, the evaluation method of sheet metal curvature error based on discrete point directly affects the accuracy and efficiency of sheet metal forming correction process. The evaluation process of sheet curvature error occurs after the reconstruction of the surface, and it needs to be compared with the target surface for maximum matching to reduce the error with the target curvature. In this paper, the plate is coarse matched according to a suitable feature point and feature vector, and then the particle swarm optimization algorithm based on simulated annealing is used for fine matching to minimize the error, and finally the evaluation result of rolling bending is displayed, which provides data support for the secondary automatic processing. The particle swarm optimization algorithm based on simulated annealing can reduce the matching error, shorten the processing time of workers, improve the working efficiency of shipyards, and has a good application prospect in steel plate processing.
Blockchain interoperability facilitates the transmission and interaction of data across different blockchain networks, ensuring communication and data exchange. Blockchain federated learning is a method that promotes the blockchain technology to address the challenge of information silos. Participants in blockchain federated learning typically store their data on private blockchains to ensure that data is shared among specific members. However, careful consideration must be given to the issue of data privacy leakage during blockchain interoperability. Malicious participants in federated learning may illicitly access private data through various attack methods, diminishing overall reliability and credibility. In order to mitigate the risk of data sharing among untrusted clients, we propose the Federated Learning-Relay Chain (FL-R) framework. By leveraging the high scalability of relay chain technology and FL, the private set intersection of distinct blockchain data within the paradigm has been achieved, ensures that only legitimate and authorized parties can participate in the federated learning process, aiming to enhance the efficiency and security of the blockchain federated learning framework. Results analysis shows that FL-R exhibits strong stability when dealing with large amounts of data. Moreover, in the same global epoch, FL-R demonstrated a 3.7% increase in accuracy compared to FedProx, ensuring the secure and reliable transmission of data within the blockchain.
The role of water consumption prediction in urban water supply system dispatch is becoming increasingly significant, and reliable daily water demand prediction models are of great importance for the construction of smart water and smart cities. In response to the problems of existing models such as automatic optimization of hyperparameters and non-stationary, non-linear water consumption data, a fresh daily water consumption forecast model is suggested in this research. This model is based on the gated recurrent unit network model, using historical water consumption data, factors affecting water consumption, and reinforcement learning dynamic adjustment of hyperparameter strategies for training. The model is evaluated using a dataset of real water supply facilities. The findings demonstrate that the suggested model is superior to conventional prediction techniques and may contribute to more efficient and sustainable water management practices.
In knowledge graph-based recommendation systems, modeling based on item features is a crucial direction. To further enhance the effectiveness of recommendation, we can study common similarities among items based on their shared attributes. The attribute information of items can assist recommendation models in understanding the structure and semantics of items. Existing models face the challenge of not considering user intent when interpreting item attribute information. In this paper, we propose the KGCIN model, which explores user preferences on the knowledge graph. By integrating the structure and semantic information of the knowledge graph and increasing the receptive field factor of entities, the KGCIN model ultimately obtains item embeddings with neighborhood information aggregation results that incorporate user preference understanding. On the other hand, user representations are obtained through information aggregation in the user intent graph. Finally, personalized recommendations for users are made by aggregating representations of user preferences and item information. Experiments in this paper are conducted on sparse dataset Last.FM and dense dataset MOOPer, respectively, validating that the proposed model outperforms other methods based on knowledge graph information fusion or user preference understanding on different datasets. According to experimental results, the model shows improvements in recommendation metrics such as precision and recall.
