Yan Cui

With production growth and improvements in production speed in modern industries, accurate anomaly detection is becoming increasingly important to improve quality inspection efficiency and help minimize production costs. Existing reconstruction-based methods achieve promising anomaly detection results in some scenarios. However, when large-scale anomalies exist, the generalization ability of these methods is limited, and it is difficult to reconstruct the anomalies effectively into normal areas, which may lead to unsatisfactory results. To address this issue, this paper proposes a novel network architecture called SeRNet, which comprises three components: a segmentation sub-network, a reconstruction sub-network, and a repair module. SeRNet addresses the challenge of large-scale anomaly reconstruction by utilizing the segmentation sub-network for pre-segmentation, the repair module for repairing the large-scale anomalies using normal images similar to the test images, and the reconstruction sub-network for processing small-scale anomalies and anomalies in the repaired splices. Additionally, this paper introduces two methods for generating pseudo-anomalies at different scales. SeRNet leverages the advantages of both the segmentation sub-network, which is effective at segmenting large-scale anomalies, and the reconstruction sub-network, which can effectively reconstruct small-scale anomalies. Experiments on the MVTec AD industrial dataset demonstrate that SeRNet delivers outstanding performance, achieving an image-level AUROC score of 99.6.

This paper proposes a new proximal iteratively reweighted nuclear norm method for a class of nonconvex and nonsmooth optimization problems. The primary contribution of this work is the incorporation of line search technique based on dimensionality reduction and extrapolation. This strategy overcomes parameter constraints by enabling adaptive dynamic adjustment of the extrapolation/proximal parameters (α_k, β_k, μ_k). Under the Kurdyka–Łojasiewicz framework for nonconvex and nonsmooth optimization, we prove the global convergence and linear convergence rate of the proposed algorithm. Additionally, through numerical experiments using synthetic and real data in matrix completion problems, we validate the superior performance of the proposed method over well-known methods.

Recommendation systems are designed to uncover users’ potential preferences and make recommendations. However, they often face challenges such as data sparsity and the cold start problem. Although the introduction of knowledge graphs has partially addressed the issue of data sparsity, the challenge of cold start has not been effectively resolved. In this article, a novel approach called Social Perception with Graph Attention Network (SPGAT) for Recommendation is proposed. In SPGAT, we aim to leverage social perception to solve the cold start effectively for more accurate recommendations. The approach utilizes a multi-layer graph attention network to aggregate user preference features from collaborative knowledge graphs and social perception graphs. By analyzing the social network of a new user, associated friend users can be identified. The interaction data of these friend users is then provided as side information to recommend to the new user. To handle one-to-many and many-to-many relations, we introduce the TransD graph embedding model, which maps different types of relations and entities to different spaces. To optimize the proposed SPGAT, self-adversarial negative sampling is utilized to implement entity and relation embedding and generate negative samples. Experimental results demonstrate that SPGAT has achieved superior performance compared to several advanced methods.

In this paper, we combine the teaching and learning situation of deaf and hard-of-hearing students in the Linear Algebra course of the Computer Science and Technology major at the Nanjing Normal University of Special Education. Based on the cognitive style of deaf and hard-of-hearing students, we apply example induction, exhaustive induction, and mathematical induction to the teaching of Linear Algebra by utilizing specific course content. The aim is to design comprehensive teaching that caters to the cognitive style characteristics of deaf and hard-of-hearing students, strengthen their mathematical thinking styles such as quantitative thinking, algorithmic thinking, symbolic thinking, visual thinking, logical thinking, and creative thinking, and enhance the effectiveness of classroom teaching and learning outcomes in Linear Algebra for deaf and hard-of-hearing students.

Unsupervised anomaly detection methods are widely used in industrial anomaly detection, primarily since there is a lack of anomalous data available for training. However, these methods still struggle to effectively detect and localize anomalies due to the diverse types of anomalies and frequent variations in their sizes and shapes. To address this challenge, we propose a new flow-based framework called Two-Hierarchy Normalizing Flow (THF) for anomaly detection and localization. THF consists of two main components: a masked flow and a constant flow. In the masked flow, we improve the traditional convolutional kernel by masking partial regions of the kernel to prevent excessive minimization of negative log-likelihood. This enhancement greatly helps in effectively detecting and localizing anomalies. The constant flow incorporates dual units and a volume-preserving flow module. The dual units consist of a characterization unit and a mixture unit. The characterization unit accurately captures both local and global feature information, while the mixture unit learns neighboring feature representations. Unlike traditional flow-based methods, the volume-preserving flow module maintains the volume of the probability distribution invariable, facilitating precise detection and localization of high-scoring anomalies. Extensive experiments conducted on multiple widely used anomaly detection datasets demonstrate that THF outperforms state-of-the-art methods in both anomaly detection and localization.

With the continuous advancement of deep learning technology, U-Net–based algorithms for image denoising play a crucial role in medical image processing. However, most U-Net-based medical image denoising algorithms typically have large parameter sizes, which poses significant limitations in practical applications where computational resources are limited or large-scale patient data processing are required. In this paper, we propose a medical image denoising algorithm called AsymUNet, developed using an asymmetric U-Net framework and a spatially rearranged multilayer perceptron (MLP). AsymUNet utilizes an asymmetric U-Net to reduce the computational burden, while a multiscale feature fusion module enhances the feature interaction between the encoder and decoder. To better preserve the image details, spatially rearranged MLP blocks serve as the core building blocks of AsymUNet. These blocks effectively extract both the local and global features of the image, reducing the model’s reliance on prior knowledge of the image and further accelerating the training and inference processes. Experimental results demonstrate that AsymUNet achieves superior performance metrics and visual results compared with other state-of-the-art methods.

Deep convolutional neural networks (CNNs) have demonstrated significant potential in enhancing image denoising performance. However, most denoising methods fuse different levels of features through long and short skip connections, easily generating a lot of redundant information, thereby weakening the complementarity of different levels of features, resulting in the loss of image details. In this paper, we propose a multi-scale detail–noise complementary learning (MDNCL) network for additive white Gaussian noise removal and real-world noise removal. The MDNCL network comprises two branches, namely the Detail Feature Learning Branch (DLB) and the Noise Learning Branch (NLB). Specifically, a loss function is applied to guide the complementary learning of image detail features and noisy mappings in these two branches. This learning approach effectively balances noise reduction and detail restoration, especially when dealing with high ratios of noise. To enhance the complementarity of features between different network layers and avoid redundant information, we designed a Feature Subtraction Unit (FSU) to capture the differences in features across the DLB network layers. Our extensive experimental evaluations demonstrate that the MDNCL approach achieves impressive denoising performance and outperforms other popular denoising methods.

Speech emotion recognition (SER) methods rely on frames to analyze the speech data. However, the existing methods typically divide a speech sample into smaller speech frames and label them with a single emotional tag, which fails to consider the possibility of multiple emotion tags coexisting within a speech sample. To deal with this limitation, we present a novel approach called self-labeling learning ensemble via DRNN and self-representation (En-DRNN-SR) for SER. This method automatically segments speech sample into speech frames, and then the deep recurrent neural network (DRNN) is applied to learn the deep features, and next the self-representation is built to get a relational degree matrix, finally the speech frames is divided into three parts using a relational degree matrix: the key emotional frames, the compatible emotional frames and the noise frames. The emotion tags of the compatible emotional frames are adaptive cyclic learned based on the key emotion frames vias the relational degree matrix, while also checking the emotion tags associated with the key compatible frames. Additionally, we introduce a new self-labeling criterion based on fuzzy membership degree for SER. To evaluate the feasibility and effectiveness of the proposed En-DRNN-SR, we conducted extensive experiments on IEMOCAP, EMODB, and SAVEE database, the proposed En-DRNN-SR obtains 69.13%, 82.83%, and 52.31% results on IEMOCAP, EMODB, and SAVEE database, which outperformed all competing algorithms. The experimental results clearly demonstrate that the proposed approach outperforms state-of-the-art SER methods, achieving superior performance on feature learning and classification.