Abstract:In today''s digital age, Nginx has emerged as the most prevalent web application server on Linux systems, securing the top position in market share. Given its critical role in ensuring the quality of service for users, optimizing the performance of Nginx servers is important. Despite the widespread deployment of Nginx servers across the two main hardware architectures, X86 and ARM, a comparative analysis of performance tuning on these architectures remains unexplored. This study aims to bridge this gap by employing automatic system parameter tuning on Nginx across these architectures, revealing the significant difference. When handling dynamic requests, the optimized performance of Nginx on X86 architecture significantly outperforms that of the ARM architecture. As a result, the optimized performance of Nginx on X86 architecture achieves a P99 latency of 515 milliseconds, which is performance improvement of 287% than that of the ARM architecture. Conversely, when processing static requests, the ARM architecture demonstrates superior performance, with a P99 latency of 220 milliseconds, resulting in a performance increase of 60% than that of X86 architecture. These findings highlight the distinct advantages of X86 and ARM architectures in handling different types of loads. It shows the significant impact of hardware architecture on optimizing Nginx’s performance. Therefore, to optimize the performance of Nginx web server, system administrators must consider the performance differences between static and dynamic requests of Nginx and the unique iterative efficiency over different hardware architectures.

Abstract:Cardiovascular disease is one of leading threats to human life and health. Tissue-engineered vascular scaffolds that can assist the regeneration/repair of disordered vessels have provided promising alternatives for cardiovascular disease treatment. However, existing tissue-engineered vascular scaffolds still confront grand challenges in interfacial adaptations, resulting in high risks of complications upon implantation and unsatisfactory translational application. Recently, tissue-engineered vascular scaffolds capable of programmed deformation have been emerging. Such scaffolds can not only dynamically adapt to three-dimensional vascular shapes with varying diameters but also orderly regulate behaviors and functions of vascular cells, offering new opportunities for addressing the grand challenges of interfacial adaptations. An overview of most-updated advances and perspectives of programmed deformed scaffolds for vascular tissue engineering will provide valuable inspirations to the development and translational applications of new generation of tissue-engineered vascular scaffolds.

Abstract:Artificial intelligence interpretability refers to the ability of people to understand and interpret the decision-making process of machine learning models. Research in this field aims to improve the transparency of machine learning algorithms, making their decisions more trustworthy and explainable. Interpretability is crucial in artificial intelligence systems, especially in sensitive and critical decision-making domains such as healthcare, finance, and law. By providing interpretability, people can better understand the reasoning behind the model''s decisions, ensuring that they are fair, robust, and ethical. In the continuously evolving field of artificial intelligence, enhancing the interpretability of models is a key step towards achieving trustworthy and sustainable AI. The article outlines the development history of interpretable artificial intelligence and the technical characteristics of various interpretability methods, with a particular focus on interpretability in the medical field. It provides a more in-depth discussion of the limitations of current methods on medical imaging datasets and proposes possible future directions for exploration.

Abstract:Over the past few decades, the rapid development and widespread adoption of internet technology have propelled humanity into the digital information age, The internet has evolved into a crucial component of human life. With the emergence of the digital lifestyle, individuals are continously generating massive amounts of digital information. Effective and convenient storage of this information is regarded as a significant challenge that needs to be overcome. In this article, we start with introducing the existing storage methods and media, and analyzing the current state of the storage field. Subsequently, we delve into the advantages, core technologies, and the potential applications of DNA as a big data storage medium in the coming days. Furthermore, we propose the future development trends and give insights into DNA-based information storage. We aim to offer new thoughts for the advancement of DNA-based data storage technology.

Abstract:As food plays an important role in people''s daily lives, a food map showing the geographical distribution of restaurants in a city is of great social value. Social media has covered every aspect of people''s lives; therefore, social media data provides a wealth of data to support automatic cartography. This work proposes an automatic generation method for urban food maps driven by social media data, integrating machine learning and cartographic algorithms to realize the intelligent generation of stylized urban food maps. A visualization system of urban food maps has been developed, which is applied to four cities, Wuhan, Guangzhou, Chongqing, and Chengdu, for case studies. The results show the effectiveness and good visual expressiveness of our method in presenting urban cuisine for cities.

Abstract:Convolutional Neural Networks (CNNs), as a quintessential representation of deep learning, are the most commonly used neural networks in tasks such as computer vision. However, convolution operations typically account for over 90% of the runtime in CNNs, becoming a bottleneck for performance. Additionally, due to the complexity of current hardware and the diversity of workloads, specific optimizations in previous work often lack performance portability. To address this, we introduce BlazerML, an open-source convolution computation library based on auto-generated code templates from TVM, capable of automatically generating high-performance convolution implementations for any input shape. BlazerML is implemented based on the Winograd algorithm, known for its high performance in fast convolution algorithms. Experimental results demonstrate that BlazerML significantly outperforms current state-of-the-art open-source libraries. On x86 CPUs, running common deep learning network forward inferences, it is faster by 1.18~2.47, 1.18~2.27, and 1.01~1.66 times compared to OnnxRuntime, MNN, and the TVM community version, respectively. On ARM CPUs, for single-layer inference of common deep learning networks, it surpasses ACL and FastConv by 1.26~6.11 and 1.04~4.28 times, respectively.

Abstract:Recommendation systems can effectively address the problem of information overload, attracting extensive attention from both academia and industry. Collaborative filtering algorithms based on graph neural networks have emerged as a widely adopted technique in recent years. These algorithms can effectively represent user and item features and learn intricate relationships between users and items. Therefore, they have become prevalent in the field of recommendation systems. In this paper, we first categorize the algorithms based on the problems that they aim to solve and then provides a comparison and analysis of representative algorithms within each category. We also summarize commonly used datasets in experiments and briefly introduce the key evaluation metrics. Finally, we discuss the challenges and potential research directions.

Abstract:Action recognition in the dark is a challenging task in practice because it is difficult to learn robust action representations from low light environments. Furthermore, there is a domain gap between dark scenes and the data used by traditional pretrained models, which results in suboptimal results with the traditional pretrain-finetune approach, and pretraining from scratch is costly. To address this issue, a domain-adaptive pretraining method is proposed to improve action recognition performance in the dark environments. The method integrates an external vision enhancement model for de-darkening to introduce critical knowledge for dark scene processing. It also employs a cross-domain self-distillation framework to reduce the domain gap of visual representations between illuminated and dark scenes. Through extensive experiments in various dark environment action recognition settings, the proposed approach can achieve a Top-1 accuracy of 97.19% on the dark dataset of fully supervised action recognition. In the source-free domain adaptation on the Daily-DA dataset, the accuracy can be improved to 49.11%. In the multi-source domain adaptation scenario on the Daily-DA dataset, the Top-1 accuracy can reach 54.63%.

Abstract:The growing contradiction between the exponential increase in data amount and the limited storage capacity of existing media is becoming increasingly evident, necessitating the development of new types of media to address this issue. Due to its ultra-high density, low energy consumption and long lifetime for data storage, DNA has attracted significant attention as an emerging storage medium, particularly for massive “cold data”, with the potential to replace current storage methods. In the process of data storage, the effective preservation of DNA plays a crucial role, directly impacting the quality of storage density, stability, storage time, as well as data writing and reading. Due to the limited information available on DNA preservation techniques in the current literature, this paper provides an overview of current research progress and strategies in DNA preservation technology for data storage, discusses the difficulties and challenges faced when applying existing preservation techniques in DNA data storage, and presents prospects for the implementation of DNA data storage.

Abstract:Effectively transferring knowledge from pre-trained models to downstream video understanding tasks is an important topic in computer vision research. Knowledge transfer becomes more challenging in open world due to poor data conditions. Many recent multimodal pre-training models are inspired by natural language processing and perform transfer learning by designing prompt learning. In this paper, we propose an LLM-powered domain context-assisted open-world action recognition method that leverages the open-world understanding capabilities of large language models. Our approach aligns visual representation with multi-level descriptions of human actions for robust classification, by enriching action labels with contextual knowledge in large language model. In the experiments of open-world action recognition with fully supervised setting, we obtain a Top-1 accuracy of 71.86% on the ARID dataset, and an mAP of 80.93% on the Tiny-VARIT dataset. More important, our method can achieve Top-1 accuracy of 48.63% in source-free video domain adaptation and 54.36% in multi-source video domain adaptation.