With the rapid development of Vehicle-Road-Network integration, trajectory prediction has played a critical role in traffic efficiency enhancement and driving safety. Feasible trajectory prediction can be achieved by deep learning and big data methods. However, distributed data in different vehicles leads to Isolated Data Island issue, which becomes a challenge for effective model updating and safe data sharing. In this paper, we propose a novel algorithm for Internet of Vehicle (IoV) trajectory prediction and privacy protection under the secure federated learning (FL) scheme. It utilizes graph neural network (GNN) to effectively represent vehicle nodes and their feature information, which satisfies communication efficiency and privacy protection simultaneously with FL, named FedGNN. First, update vehicle feature as embedding with GNN local model in Client-Side for trajectory prediction. Then, FL Server-Side introduces a secure aggregation approach to generate global model to protect vehicle-to-vehicle (V2V) communication privacy. Simulation on NGSIM and highD datasets and comparison with baseline algorithms verify that FedGNN has advantages in both trajectory prediction performance and timeliness.

ABSTRACT The environment of fully mechanized mining face is complex, and most coal targets are small, dense and prone to overlap or occlude. Existing object detection algorithms are difficult to extract features and has poor generalization ability, which makes it difficult to apply to the scene of fully mechanized mining face. To alleviate these problems, we propose a novel patch‐aware and cross‐scale feature fusion network MBN‐YOLOv8 for large coal detection in fully mechanized mining face, which consists of three main innovations: Bidirectional cross‐scale connections and weighted feature fusion network (BiFPN), Multi‐branch patch‐aware attention module (MPA), and Normalized Wasserstein distance loss function (NWDCIoU). The MPA employs multi‐branch feature extraction, patch awareness and an attention mechanism to capture small‐scale feature details while enhancing large‐scale feature information. The BiFPN utilizes a bidirectional path feature transfer mechanism and a weighted feature fusion strategy to ensure more comprehensive and detailed cross‐layer transfer and fusion of feature information. The NWDCIoU loss function measures the similarity of the derived Gaussian distribution using normalized wasserstein distance, making it easier for the network to detect scale‐insensitive small targets. Experimental results on the DsLMF+ dataset indicate that the proposed MBN‐YOLOv8 achieves state‐of‐the‐art performance, with [email protected] of 0.872 and precision of 0.825, significantly outperforming nine other state‐of‐the‐art baseline models.

Recent studies have noted an intriguing phenomenon termed Neural Collapse, that is, when the neural networks establish the right correlation between feature spaces and the training targets, their last-layer features, together with the classifier weights, will collapse into a stable and sym-metric structure. In this paper, we extend the investigation of Neural Collapse to the biased datasets with im-balanced attributes. We observe that models will easily fall into the pitfall of shortcut learning and form a biased, non-collapsed feature space at the early period of training, which is hard to reverse and limits the generalization capability. To tackle the root cause of biased classification, we follow the recent inspiration of prime training, and propose an avoid-shortcut learning framework without ad-ditional training complexity. With well-designed shortcut primes based on Neural Collapse structure, the models are encouraged to skip the pursuit of simple shortcuts and nat-urally capture the intrinsic correlations. Experimental re-sults demonstrate that our method induces better conver-gence properties during training, and achieves state-of-the-art generalization performance on both synthetic and real-world biased datasets.

Spectrum maps are of crucial importance for realizing efficient spectrum management in the sixth-generation (6G) wireless communication networks. However, existing spectrum map construction schemes mainly depend on spatial interpolation or just simply exploit the frequency correlation and cannot accurately construct the spectrum map when measurement data at the target frequency are not available. To overcome this challenge, we propose two accurate spectrum map construction schemes using different intelligent frequency-spatial reasoning methods. The frequency correlation among different spectrum maps at different frequencies is fully exploited to construct highly accurate spectrum maps of the frequencies without spectrum data by combining the joint frequency-spatial spectrum representation method with deep learning data-driven techniques. Specifically, a joint three-dimensional spectrum representation model is established and both a novel autoencoder network and a novel conditional generative adversarial network suitable for processing the three-dimensional spectrum data are proposed to realize the intelligent frequency-spatial reasoning. Simulation results demonstrate that our proposed schemes are superior to the benchmark schemes in terms of the spectrum map construction accuracy. Moreover, simulation results demonstrate that our proposed neural networks have a fast convergence speed, which achieves a better tradeoff between the computation efficiency and the construction accuracy.

Spectrum map is of crucial importance for realizing efficient spectrum management in the sixth-generation (6G) wireless communication networks. However, the existing spectrum map construction schemes mainly depend on spatial interpolation and cannot construct the spectrum map when the measurement data of the target frequency are not obtained. In order to overcome this challenge, an accurate spectrum map construction scheme is proposed by using an intelligent frequency-spatial reasoning approach. The frequency correlation among different spectrum maps at different frequencies is fully exploited to construct the highly accurate spectrum maps of the frequencies without spectrum data. A novel autoencoder adapting to the three-dimensional (3D) spectrum data is proposed. Simulation results demonstrate that our proposed scheme is superior to the benchmark schemes in terms of the construction accuracy. Moreover, it is shown that our proposed autoencoder network has a fast convergence speed.

As China promotes the continuous process of opening-up and expands its influence, people around the world now accept and recognize Chinese culture. An increasing number of foreign products and brands enrich the life of Chinese people, and Chinese brands and products are also expected to shine brilliantly on the world stage and tap into the world market, so as to grace the world with their unique charm. Based on the tableware designed by Chinese original brands, this paper analyzes patterns, market and existing products of current tableware, provides design and competition strategies for Chinese tableware brands by taking the example of tableware pattern design.

Embedding has emerged as an important approach to prediction, inference, data mining and information retrieval based on knowledge bases and various embedding models have been presented. Most of these models are "typeless," namely, treating a knowledge base solely as a collection of instances without considering the types of the entities therein. In this paper, we investigate the use of entity type information for knowledge base embedding. We present a framework that augments a generic "typeless" embedding model to a typed one. The framework interprets an entity type as a constraint on the set of all entities and let these type constraints induce isomorphically a set of subsets in the embedding space. Additional cost functions are then introduced to model the fitness between these constraints and the embedding of entities and relations. A concrete example scheme of the framework is proposed. We demonstrate experimentally that this framework offers improved embedding performance over the typeless models and other typed models.