With the rapid growth of multi-modal content on the Web, robust vision-language models are essential for semantic understanding and classification of web images under diverse and dynamic contexts, supporting Web applications such as multimedia search and recommendation. Prompt learning has proven effective for enhancing vision-language models in semantic image classification tasks. However, previous methods often suffer from poor generalization: the learned prompts tend to overfit the base classes seen during training, leading to poor performance on unseen classes and under distribution shifts. This issue is especially challenging in Web-scale data, where new classes emerge and distributions shift dynamically. To address these limitations, we propose PLIKD, a novel prompt learning method that integrates instance-aware knowledge distillation for robust Web-scale semantic image classification. Specifically, PLIKD introduces an instance-aware knowledge extraction module, which leverages multi-modal large language models through a step-by-step strategy to extract external knowledge for each image instance. To incorporate this extracted knowledge, PLIKD further introduces an instance-aware knowledge distillation module, which consists of two key steps: (1) a dual-teacher strategy for robust and informative knowledge distillation, and (2) fine-grained cross-modal alignment via Smooth and Sparse Optimal Transport. Extensive experiments demonstrate that PLIKD significantly improves generalization to both seen and unseen classes, and remains robust under distribution shifts, outperforming existing state-of-the-art methods on Web-scale semantic image classification.

Manual annotation for crowd counting remains labor-intensive and costly. Although existing semi-supervised methods partially alleviate this burden, they still face significant challenges regarding the quality of generated pseudo-labels and the utilization of unlabeled data. To address these issues, we propose a novel semi-supervised crowd counting framework, called Point-Adaptive Teacher (PAT). This framework integrates Adaptive Soft Threshold (AST) and contrastive learning to enhance pseudo-label quality and effectively leverage unlabeled data. Specifically, we employ the Swin Transformer as the backbone and develop Swin-P2PNet, which captures global contextual information through hierarchical window attention, improving the accuracy of pseudo-labels. Additionally, we design the AST that dynamically adjusts the sample loss weight by combining confidence and uncertainty predictions, thereby alleviating the effect of noise in pseudo-labels. Finally, we introduce a contrastive learning strategy requiring no extra parameters. This strategy enhances the model’s ability to learn latent representations from unlabeled data. Extensive experiments have been conducted on three public datasets, namely ShanghaiTech, JHU-Crowd++, and UCF-QNRF. The results demonstrate that our method achieves performance comparable to state-of-the-art methods.

Next Point-of-Interest (POI) recommendation has become a crucial task in Location-Based Social Networks (LBSNs), which provide personalized recommendations by predicting the user's next check-in locations. Commonly used models including Recurrent Neural Networks (RNNs) and Graph Convolutional Networks (GCNs) have been widely explored. However, these models face significant challenges, including the difficulty of capturing the hierarchical and tree-like structure of POIs in Euclidean space and the sparsity problem inherent in POI recommendations. To address these challenges, we propose a Hyperbolic Variational Graph Auto-Encoder (HVGAE) for next POI recommendation. Specifically, we utilize a Hyperbolic Graph Convolutional Network (Hyperbolic GCN) to model hierarchical structures and tree-like relationships by converting node embeddings from euclidean space to hyperbolic space. Then we use Variational Graph Auto-Encoder (VGAE) to convert node embeddings to probabilistic distributions, enhancing the capture of deeper latent features and providing a more robust model structure. Furthermore, we combine the Mamba4Rec recommender and Rotary Position Embedding (RoPE) and propose Rotary Position Mamba (RPMamba) to effectively utilize POI embeddings rich in sequential information, which improves the accuracy of the next POI recommendation. Extensive experiments on three public datasets demonstrate the superior performance of the HVGAE model.

With the rapid development of the mobile internet, there is an increasing demand for quick access to effective data. Consequently, more research is focusing on data processing and optimization of recommendation systems in edge computing environments. However, in traditional edge computing environments, recommendation systems typically depend on frequent data query interactions among all edge servers to obtain results, which increases time delays. This issue is further exacerbated by the need to process large amounts of data within edge storage systems. To address this challenge, we propose an efficient recommendation method based on data allocation. Specifically, during the data allocation process, we first extract similar features of users, ensuring that the characteristics of some user data align as closely as possible with the overall user feature distribution. Then, using an improved consistent hashing algorithm, we achieve a uniform data layout, allowing the recommendation system to efficiently and accurately provide recommendations by querying data from the nearest edge server to the user. Finally, extensive experiments on real datasets show that our method significantly reduces time delays and improves the accuracy of recommendation results.

Recently, patch-based transformer methods have demonstrated strong effectiveness in time series forecasting. However, the complexity of self-attention imposes demands on memory and compute resources. In addition, though patches can capture comprehensive temporal information while preserving locality, temporal information within patches remains important for time series prediction. The existing methods mainly focus on modeling long-term dependencies across patches, while paying little attention to the short-term dependencies within patches. In this paper, we propose the Global and Local Frequency-domain Network (GLFNet), a novel architecture that efficiently learns global time dependencies and local time relationships in the frequency domain. Specifically, we design a frequency filtering layer to learn the temporal interactions instead of self-attention. Then we devise a dual filtering block consisting of global filter block and local filter block which learns the global dependencies across patches and local dependencies within patches. Experiments on seven benchmark datasets demonstrate that our approach achieve superior performance with improved efficiency.

The provision of privacy-preserving recommendations for geological tourist attractions is an important research area. The historical check-in data collected from location-based social networks (LBSNs) can be utilized to mine their preferences, thereby facilitating the promotion of the geological tourism industry. However, such check-ins often contain sensitive user information that poses privacy leakage risks. To address this issue, some methods have been proposed to develop privacy-preserving point-of-interest (POI) recommendation systems. These methods commonly rely on either perturbation-based or federated learning techniques to protect users’ privacy. However, the former can hinder preference capture, while the latter remains vulnerable to privacy breaches during the parameter-sharing process. To overcome these challenges, we propose a novel privacy-preserving POI recommendation model that incorporates users’ privacy preferences based on a simplified graph convolutional neural network. Specifically, we employ a generative model to create a subset of POIs that reflect users’ preferences but do not reveal their private information, and then we design a simplified graph convolutional network to analyze the high-order connectivity between users and POIs that are privacy-preserving. The resulting model enables efficient POI recommendation under strict privacy protection, which is particularly relevant to geological tourism. Experimental results on two public datasets demonstrate the effectiveness of our proposed approach.

Inkjet printing technology can make tiles with very rich and realistic patterns, so it is widely adopted in the ceramic industry. However, the frequent nozzle blockage and inconsistent inkjet volume by inkjet printing devices, usually leads to defects such as stayguy and color blocks in the tile surface. Especially, the stayguy in complex pattern is difficult to identify by naked eyes due to it is easily covered by complex patterns and becomes invisible, this brings great challenge to tile quality inspection. Nowadays, the machine learning is employed to address the issues. The existing machine learning methods based on hand-crafted features are capable of stayguy detection of the tiles with a simple pattern, but not applicable for complex patterns due to the interference of pattern in feature extraction. The emerging deep-learning-based methods have the potential to be applied for stayguy detection with complex patterns, but cannot achieve real-time detection due to high complexity. In this paper, a lightweight hand-crafted feature enhanced convolutional neural network (named HFENet) is proposed for rapid defect detection of tile surface. First, we perform data enhancement on the original image by global histogram equalization and image addition. Second, for the special shape of stayguy which is usually vertical, we embed the extended vertical edge detection operator (Prewitt) as convolution kernel into HFENet to extract the hand-crafted vertical edge features of the test image and eliminate the interference of complex pattern in the feature extraction. Third, the 5 × 1 asymmetric convolution kernel with a dilation rate of 2 is used to improve the utilization of convolution kernel and reduce the complexity of the model. Fourth, to reach the real-time requirements, a memory access cost-aware design is proposed, which can orchestrate the number of shallow convolution layers and deep convolution layers in feature extraction. The experiments were performed on the ceramic tile image data set captured by high-resolution industrial cameras in ceramic tile production line. Experimental results show that the HFENet outperforms the state-of-the-art semantic segmentation networks (i.e., UNet, FCN-8s, SegNet, DeepLabV3+, etc.) and lightweight networks (i.e., ShuffleNet, MobileNet, and SqueezeNet). All the code and data are available at a GitHub repository (https://github.com/RobotvisionLab/HFENet).