In Natural Language Processing (NLP), dealing with underrepresented topics is challenging, especially in unsupervised tasks where clustering might not adequately capture minority topics. To tackle this challenge, our ...
详细信息
Video super-resolution is a fundamental task aimed at enhancing video quality through intricate modeling techniques. Recent advancements in diffusion models have significantly enhanced image super-resolution processin...
详细信息
Video super-resolution is a fundamental task aimed at enhancing video quality through intricate modeling techniques. Recent advancements in diffusion models have significantly enhanced image super-resolution processing capabilities. However, their integration into video super-resolution workflows remains constrained due to the computational complexity of temporal fusion modules, demanding more computational resources compared to their image counterparts. To address this challenge, we propose a novel approach: a Frames-Shift Diffusion Model based on the image diffusion models. Compared to directly training diffusion-based video super-resolution models, redesigning the diffusion process of image models without introducing complex temporal modules requires minimal training consumption. We incorporate temporal information into the image super-resolution diffusion model by using optical flow and perform multi-frame fusion. This model adapts the diffusion process to smoothly transition from image super-resolution to video super-resolution diffusion without additional weight parameters. As a result, the Frames-Shift Diffusion Model efficiently processes videos frame by frame while maintaining computational efficiency and achieving superior performance. It enhances perceptual quality and achieves comparable performance to other state-of-the-art diffusion-based VSR methods in PSNR and SSIM. This approach optimizes video super-resolution by simplifying the integration of temporal data, thus addressing key challenges in the field.
In today’s digital era, blockchain, big data, cloud computing and other information technologies are booming, providing a powerful impetus for the rapid progress of human society. At the same time, the importance of ...
详细信息
In the process of refining Knowledge Graphs (KGs), new entities emerge, and old entities evolve, which usually updates their attribute information and neighborhood structures. This results in a distribution shift prob...
详细信息
In the process of refining Knowledge Graphs (KGs), new entities emerge, and old entities evolve, which usually updates their attribute information and neighborhood structures. This results in a distribution shift problem for entity features in the embedding space during graph representation learning. Most of existing inductive knowledge graph embedding methods focus mainly on the representation learning of new entities, neglecting the negative impact caused by distribution shift of entity features. In this paper, we use the skill of mean and variance reconstruction to develop a novel inductive knowledge graph embedding model named EDSU for processing the shift of entity feature distribution. Specifically, by assuming that the embedding feature of entity follows multivariate Gaussian distribution, the reconstruction combines the distribution characteristics of components in an entity embedding vector with neighborhood structure information of a set of entity embedding vectors, in order to alleviate the deviation of data information between intra-entity and inter-entity. Furthermore, the connection between the entity features distributions before and after the shift is established, which guides the model training process and provides an interpretation on the rationality of such handling distribution shift in view of distributional data augmentation. Extensive experiments have been conducted and the results demonstrate that our EDSU model outperforms previous state-of-the-art baseline models on inductive link prediction tasks.
The skip list is a popular in-memory index in modern database systems. It maintains multiple levels of lists, which makes it efficient in traversing sorted data. In addition, it is flexible in inserting and deleting d...
详细信息
The skip list is a popular in-memory index in modern database systems. It maintains multiple levels of lists, which makes it efficient in traversing sorted data. In addition, it is flexible in inserting and deleting data, while avoiding the restructuring overhead of tree-based structures. However, there are considerable challenges in the conventional skip list design. First, the linked list structure has a drawback in utilizing microarchitecture features such as cache, pipeline, and SIMD (Single Instruction Multiple Data) capability. Second, the skip list randomly selects the level of a new node. That is, the skip list runs based on probability rather than data distribution, which can lead to suboptimal lookup performance. Unlike balanced tree structures, the worst-case lookup performance of a skip list remains O(n). This paper proposes a new data structure called DASL (Deterministic Arrayed Skip List). It follows the algorithm of the skip list, but seamlessly integrates the array and devises a new deterministic raise operation in order to obtain flexibility, microarchitecture-friendliness, and reduced tail latency. In specific, a node in DASL consists of an array structure with multiple elements instead of a single element, taking advantage of the array within a list structure. Additionally, the raise operation is conducted deterministically instead of probabilistically, allowing data to be more balanced in multiple lists. Furthermore, we devise two optimization techniques, utilization-based adaptive intra-node search and uneven split operation. Experimental results with various synthetic and real-world workloads demonstrate that DASL outperforms other state-of-the-art in-memory indexes, including skip list, B+tree, and ART.
Visual tracking is a complex and crucial problem in computer vision with numerous real-world applications, including surveillance, autonomous vehicles, and augmented reality. To tackle the challenges associated with t...
详细信息
Visual tracking is a complex and crucial problem in computer vision with numerous real-world applications, including surveillance, autonomous vehicles, and augmented reality. To tackle the challenges associated with tracking performance, this paper presents Learning Disruptor-Aware Channel Selection and Reliability with Target Regularization (DACSR). First, DACSR enhances tracking robustness in challenging scenarios by adaptively selecting visual channels based on their resistance to disruptions through an intelligent disruption-aware channel selection mechanism. Second, it improves predictive accuracy and reliability by integrating a channel stability-aware normalization method, which highlights stable channels while suppressing misleading ones during optimization. Third, this study incorporates target regularization techniques using deep neural networks to capture target-specific characteristics beyond spatial considerations, further refining the tracking process. By leveraging the representational power of deep neural networks, DACSR effectively models complex target attributes, leading to improved tracking accuracy. Finally, we validate the efficiency and robustness of the proposed method through extensive experiments on benchmark datasets, demonstrating its superior performance over existing approaches in challenging tracking scenarios.
The impact of the harmful gas NO on the atmospheric environment cannot be ignored, so it is necessary to develop efficient gas sensors to detect and absorb NO at room temperature. In this work, we systematically explo...
详细信息
The impact of the harmful gas NO on the atmospheric environment cannot be ignored, so it is necessary to develop efficient gas sensors to detect and absorb NO at room temperature. In this work, we systematically explored the electron transport properties and gas sensing properties of Fe3GeTe2, In2Se3 monolayers and its multiferroic van der Waals (vdW) heterostructures of Fe3GeTe2/In2Se3 for detecting NO by using density functional theory and nonequilibrium Green's function methods. The calculated adsorption energies, electron localisation functions, band structures and density of states indicate that its a chemisorption for the adsorption of NO on Fe3GeTe2 and a physisorption on In2Se3, while the specific adsorption style is highly dependent on the stacking and molecular adsorption sites of different Fe3GeTe2/In2Se3 structures. In addition, the pristine Fe3GeTe2 monolayer and Fe3GeTe2/In2Se3 based nanodevices realized a significant anisotropy for electron transport along the zigzag and armchair directions, and their anisotropic maximum current ratios in the zigzag to armchair directions were 1.90 and 1.89. Meanwhile, the NO sensitivity ratio of Fe3GeTe2 and Fe3GeTe2/In2Se3 based gas sensors can be up to 79 % and 107 %, respectively. This highlights a notably superior gas-sensitive performance of the Fe3GeTe2/In2Se3 heterostructure compared to the Fe3GeTe2 monolayer gas devices. These findings provide valuable references for the application of multiferroic heterostructures in the field of gas sensors.
Reliable detection of traffic signs plays a critical role in autonomous vehicles and driver-assistance technologies, especially when operating under diverse and complex real-world conditions. This study investigates t...
详细信息
ISBN:
(数字)9798350357509
ISBN:
(纸本)9798350357516
Reliable detection of traffic signs plays a critical role in autonomous vehicles and driver-assistance technologies, especially when operating under diverse and complex real-world conditions. This study investigates the effectiveness of various YOLO-based object detection frameworks, ranging from YOLOv5 to the newly proposed YOLOv11, in addressing the task of traffic sign recognition. The dataset includes varying class configurations and challenging conditions such as low-light and noise, enabling a comprehensive analysis of model robustness. YOLOv11 outperformed prior versions, achieving the highest mAP of 0.961, establishing it as a state-of-the-art (SOTA) solution. The study also conducted ablations to analyze the impact of batch size, data augmentation, and adverse conditions on model performance. Notably, YOLOv11 demonstrated strong resilience to noise and low-light scenarios, maintaining high accuracy even under dense conditions. Comparisons with existing models further validated its superiority in terms of precision and consistency. These results highlight YOLOv11’s potential for real-world traffic sign detection tasks, where reliability is critical.
Trusted computing(TC)is an emerging technology to enhance the security of various computing platforms by a dedicated secure chip(TPM/TCM),which is widely accepted by both the industrial and academic *** paper attempts...
详细信息
Trusted computing(TC)is an emerging technology to enhance the security of various computing platforms by a dedicated secure chip(TPM/TCM),which is widely accepted by both the industrial and academic *** paper attempts to sketch the evolution of TC from the view of our theoretical and engineering *** theory,we focus on protocol design and security *** have proposed the first ECDAA protocol scheme based on q-SDH assumption,which highlights a new way to design direct anonymous attestation *** technical evolution,we discuss the key technologies of trust chain,trusted network connection and TC testing and *** break through several key technologies such as trusted boot,OS measurement and remote attestation,and implement a TC system from TPM/TCM to *** also design and implement a testing and evaluation system of TC platform,which is the first one put into practical application in ***,with the rapid development of cloud computing and mobile applications,TC is moving toward some new directions,such as the trust in cloud and mobile environments,new TPM standard,and flexible trust execution environment trust establishment method.
Natural language in social media is often ambiguous, carrying diverse perspectives that are critical for opinion analysis. With vast amounts of unstructured data, particularly on platforms like Twitter, it is impracti...
详细信息
ISBN:
(数字)9798331506995
ISBN:
(纸本)9798331507008
Natural language in social media is often ambiguous, carrying diverse perspectives that are critical for opinion analysis. With vast amounts of unstructured data, particularly on platforms like Twitter, it is impractical to manually analyse relevant insights. This rapid data growth has highlighted the need for intelligent filtering and mining techniques in decision support systems to help organizations build brand reputation, foster online communities, and enhance profitability. Multi-Emotion detection is key for capturing implicit user opinions and specific emotional states; however, challenges such as text bipolarity, limited representative feature vectors, and inadequate annotated lexicons often hinder accuracy. This research aims to improve multi-emotion detection accuracy by extracting emotion implied tokens and converting them into feature vectors that capture semantic relevance. Using annotated lexicons, word embeddings, unigrams, and term frequency, these vectors are trained across five machine learning models, which are Naïve Bayes (NB), Logistic Regression (LR), Support Vector Machine (SVM), K- nearest neighbor (KNN), and Multi- layer Percepton (MLP). Finally, two ensemble techniques are applied to refine model outputs, achieving an improvement accuracy over previous work worked. The proposed system, evaluated on a real dataset, achieves a hamming score of 0.55 and an average F1 score of 0.68, demonstrating its potential for effective multi-label emotion detection.
暂无评论