Software vulnerabilities are critical weaknesses that compromise the security of computer systems, making them susceptible to attacks may lead to data breaches, system crashes or even more severe security incidents. Therefore, accurately and efficiently detecting software vulnerabilities has become a central research focus in the field of computer security. Although contemporary deep learning-based vulnerability detection approaches have made progress, they are often limited by single code representations and fail to fully capture the complementary nature of code semantics and structural information. This research introduces an innovative method for software vulnerability detection, termed VDPPM (Vulnerability Detection via Path Representations and Pretrained Model), which effectively enhances code semantic analysis and vulnerability detection accuracy. VDPPM integrates the path representations extracted from abstract syntax tree, control flow graph and program dependency graphs, leverages the SimCodeBERT model optimized through contrastive learning framework SimCSE to enhance the model's ability to capture vulnerability features. In the experiments, first, three types of code representations are extracted from the source code and are used to construct a corpus by deriving path representations for the training of Doc2vec model, thus generating general-purpose embedding models, converting path sequences into vector representations. Subsequently, a pretrained CodeBERT model is integrated, which, after being trained under the contrastive learning framework, gains increased precision in capturing deep semantic features within the code. Finally, by combining vector embeddings from Doc2vec and SimCodeBERT, high-quality code representations are constructed to perform vulnerability detection. Experimental results demonstrate that, across multiple publicly available benchmark datasets for vulnerability detection tasks, VDPPM outperforms the existing mainstream methods with significant improvements in several performance metrics. This convincingly validates the effectiveness and superiority of the proposed method.

General Matrix Multiplication (GEMM) is one of the most important operations in linear algebra, serving as the backbone for numerous applications in machine learning, scientific computing, and signal processing. In particular, FP16 batch GEMM has become a core operation in deep learning frameworks due to its efficiency in training and inference. However, current implementations on AMD GPUs (e.g., CDNA/MI200 architectures with Matrix Cores) suffer from suboptimal memory access and low compute utilization, limiting performance in high-throughput scenarios. Therefore, this paper proposed a GPU optimization scheme for half-precision batch GEMM (HGEMM). In terms of blocking strategy, it allocates equal memory access and computational loads to threads based on input matrix sizes, while enabling each thread to compute multiple matrix multiplications to improve arithmetic unit utilization. For memory access optimization, it trades redundant data reads for uniform memory access patterns per thread to facilitate compiler optimization, ensuring overlapping of memory and computation time. For extremely small-batch HGEMM with matrix dimensions smaller than 16, the proposed method employs a 4 × 4 × 4 Matrix Core and its corresponding tiling scheme to enhance memory performance while reducing computational resource wastage, and provides the option of whether to use shared memory to achieve the highest performance. This paper compares the performance of this scheme with two operators of rocBLAS on the AMD GPU MI210 platform. The results show that the average performance of this scheme on AMD GPU MI210 is 4.14 times that of rocBLASHGEMMBatched and 4.96 times that of rocBLASGEMMExBatched. For extremely small-batch HGEMM, the average performance is 18.60 times that of rocBLASHGEMMBatched and 14.02 times that of rocBLASGEMMExBatched.

In recent years, with the increasing number of smart contracts and the increasing economic losses caused by contract loopholes, the security of smart contracts has attracted more and more attention. The vulnerability detection method based on deep learning can solve the problems of low detection efficiency and insufficient accuracy of the early traditional smart contract vulnerability detection method. However, most of the existing deep learning-based vulnerability detection methods directly use smart contract source code, opcode sequence or bytecode sequence as the input of the deep learning model. This fact will weaken the effective information due to the introduction of too much invalid information. To solve this problem, this paper proposed a smart contract vulnerability detection method based on capsule network and attention mechanism. Considering the execution timing information of the program, the study extracted key operation code sequence of the smart contract as the source code feature. Then a hybrid network structure of capsule network and attention mechanism was used for training. The capsule network extracts the context information of the smart contract and the connection between the part and the whole; while the attention mechanism is used to assign different weights to different opcodes according to their importance. The experimental results show that the F₁ score and accuracy of the algorithm proposed in this paper in the smart contract data set are 94.48% and 97.15%, indicating that this algorithm is superior to other detection methods in performance.

With the continuous development of digital image processing technology, image tampering is no longer limited to a single method such as image splicing, rather the traces of malicious tampering are concealed in the post processing through the image editing software. This new development leads to poor results of traditional image forgery detection algorithms and the tampering localization methods based on deep learning. Aiming at the problem of low accuracy of existing image tampering algorithms, this study proposed an end-to-end image tampering location network based on multi-scale visual Transformer. The network combines a transformer and a convolutional encoder to extract the feature difference between the tampered area and the non-tampered area. Multi-scale visual Transformer models the spatial information of image block sequences of different sizes, so that the network can adapt to tampered areas of various shapes and sizes. Experimental results show that the F₁ and AUC scores of the proposed algorithm in the CASIA and NIST2016 test sets are 0.431、0.877、0.728 and 0.971, respectively. This shows that the performance of the new algorithm is significantly better than that of the existing mainstream algorithms. Moreover, the proposed algorithm is robust against JPEG compression attacks.

Fast Fourier transform (FFT) algorithm finds widespread application in scientific computing and related fields. To fully leverage the computational power of the GPU and further enhance the performance of FFT calculations, this paper proposed a high-performance multi-dimensional FFT computation scheme based on the Matrix Core for the matrix form of Stockham FFT. In terms of computational optimization, this scheme utilizes Matrix Core to accelerate matrix multiplications in FFT computation while leveraging compiler intrinsic instructions to perform small-grained matrix multiply-accumulate operations, enabling Matrix Core to support FFT computations of more sizes. To minimize memory access, the proposed scheme directly performs matrix element-wise multiplication operations in the registers according to the distribution pattern of Matrix Core's data across thread registers. It also mitigates bank conflicts by reordering data in shared memory, adopts a double-buffering strategy to alleviate access bottlenecks, and proposes an efficient matrix transposition strategy to accelerate multidimensional FFT computations. In this paper, the proposed scheme was compared to the well-known high-performance FFT computation libraries rocFFT and VkFFT on the AMD MI250 GPU platform. The results demonstrate that the proposed scheme outperforms rocFFT and VkFFT in terms of average computational performance for 1-dimensional, 2-dimensional, and 3-dimensional FFTs on the AMD MI250 GPU platform. For 3D FFT calculation, this method has an average performance that is 1.5 times faster than rocFFT and 2.0 times faster than VkFFT, demonstrating significant performance improvements.