The mining and exploitation of security vulnerabilities have always been the focus of offensive and defensive confrontations. In recent years, with the application of technologies such as fuzzing in vulnerability mini...
详细信息
The management of Internet of Things (IoT) devices is becoming increasingly complex. One of the reasons is that IoT device manufacturers are different, and there are different degrees of heterogeneity in service, tech...
详细信息
With the continuous advancement of modern processor architectures, SIMD (Single Instruction Multiple Data) vectorization has become one of the key techniques for improving program performance. Loop vectorization, as a...
详细信息
ISBN:
(纸本)9798400713613
With the continuous advancement of modern processor architectures, SIMD (Single Instruction Multiple Data) vectorization has become one of the key techniques for improving program performance. Loop vectorization, as a core technique of SIMD optimization, can significantly enhance the execution efficiency of loop code with data parallelism. However, in practical applications, many loops contain system calls (such as printf), which often introduce side effects or uncertainties in control flow. This leads traditional compilers to adopt a conservative strategy during loop vectorization, avoiding vectorizing the entire loop. While this strategy helps prevent potential execution errors, it also leaves many loops with vectorization potential under-optimized. To address this issue, this paper proposes a loop vectorization optimization technique for loops containing system calls. By introducing directive-based guidance and appropriate handling mechanisms, the proposed method allows the compiler to identify and effectively process loops with system calls, enabling successful vectorization. Specifically, programmers can guide the compiler to selectively vectorize these loops as needed, maximizing performance improvements while ensuring program correctness. By precisely controlling which parts of the code can be vectorized and which must preserve their original execution order, this method effectively solves the problem of non-vectorizable loops with system calls, significantly improving program execution efficiency. To evaluate the effectiveness of the proposed method, we designed 24 test cases containing system calls. Experimental results show that, the highest optimization level (O3) of the compiler, the proposed method achieved an average speedup of 1.4677×. Further analysis indicates that as the ratio of computational instructions to system call instructions increases, the speedup generally rises, despite some fluctuations.
Graph few-shot learning aims to predict well by training with very few labeled data. Meta learning has been the most popular solution for few-shot learning problem. However, transductive linear probing shows that fine...
详细信息
ISBN:
(数字)9798350355925
ISBN:
(纸本)9798350355932
Graph few-shot learning aims to predict well by training with very few labeled data. Meta learning has been the most popular solution for few-shot learning problem. However, transductive linear probing shows that fine-tuning a simple linear classification head after a pretrained graph neural networks can outperforms most of the sophisticated-designed graph meta learning algorithms. Therefore, in the paper, we propose a meta transductive linear probing methods named Meta-TLP to incorporate the advantages of graph self-supervised and graph meta learning model. Specifically, the graph neural network is firstly pretrained with graph contrastive learning methods. Then we design an unsupervised meta training task construction methods to require meta tasks without relying on labeled data. Finally, we meta training the linear classification head on the meta training tasks to learn to fast adopt to novel classes. Experiment results show that our model can perform better than TLP on three real world datasets.
Detection of color images that have undergone double compression is a critical aspect of digital image *** the existence of various methods capable of detecting double Joint Photographic Experts Group(JPEG) compressio...
详细信息
Detection of color images that have undergone double compression is a critical aspect of digital image *** the existence of various methods capable of detecting double Joint Photographic Experts Group(JPEG) compression,they are unable to address the issue of mixed double compression resulting from the use of different compression *** particular,the implementation of Joint Photographic Experts Group 2000(JPEG2000)as the secondary compression standard can result in a decline or complete loss of performance in existing *** tackle this challenge of JPEG+JPEG2000 compression,a detection method based on quaternion convolutional neural networks(QCNN) is *** QCNN processes the data as a quaternion,transforming the components of a traditional convolutional neural network(CNN) into a quaternion *** relationships between the color channels of the image are preserved,and the utilization of color information is ***,the method includes a feature conversion module that converts the extracted features into quaternion statistical features,thereby amplifying the evidence of double *** results indicate that the proposed QCNN-based method improves,on average,by 27% compared to existing methods in the detection of JPEG+JPEG2000 compression.
Due to the simplicity of implementation and high threat level, SQL injection attacks are one of the oldest, most prevalent, and most destructive types of security attacks on Web-based information systems. With the con...
详细信息
Quantum NOT gates play an important role in the process of quantum information conversion. However, when the X-gate operation is executed on a real quantum computer, there is a large deviation between the actual opera...
详细信息
Convolutional neural networks (CNNs) have shown remarkable advantages in a wide range of domains at the expense of huge parameters and computations. Modern CNNs still tend to be more complex and larger to achieve bett...
详细信息
Multimodal fake news detection aims to automatically identify real or fake news, thereby mitigating the adverse effects caused by such misinformation. Although prevailing approaches have demonstrated their effectivene...
详细信息
For Internet forum Points of Interest(PoI),existing analysis methods are usually lack of usability analysis under different conditions and ignore the long-term variation,which lead to blindness in method *** address t...
详细信息
For Internet forum Points of Interest(PoI),existing analysis methods are usually lack of usability analysis under different conditions and ignore the long-term variation,which lead to blindness in method *** address this problem,this paper proposed a PoI variation prediction framework based on similarity analysis between long and short *** on the framework,this paper presented 5 PoI analysis algorithms which can be categorized into 2 types,i.e.,the traditional sequence analysis methods such as autoregressive integrated moving average model(ARIMA),support vector regressor(SVR),and the deep learning methods such as convolutional neural network(CNN),long-short term memory network(LSTM),Transformer(TRM).Specifically,this paper firstly divides observed data into long and short windows,and extracts key words as PoI of each ***,the PoI similarities between long and short windows are calculated for training and ***,series of experiments is conducted based on real Internet forum *** results show that,all the 5 algorithms could predict PoI variations well,which indicate effectiveness of the proposed *** the length of long window is small,traditional methods perform better,and SVR is the *** the contrary,the deep learning methods show superiority,and LSTM performs *** results could provide beneficial references for PoI variation analysis and prediction algorithms selection under different parameter configurations.
暂无评论