Xinyun Wu

The Minimum Hitting Set (MinHS) problem is a classic NP-hard problem in combinatorial optimization, widely applied across fields such as artificial intelligence diagnosis, bioinformatics, and network monitoring. This paper proposes a local search algorithm with Dual-evaluation mode, named DemHS (Dual-evaluation mode search algorithm), tailored for large-scale hitting set instances. The evaluation employs two types of functions: strong and weak. Strong evaluation accurately calculates the exchange gain by introducing a cross-correction term within the local neighborhood defined by an inserted vertex; meanwhile, weak evaluation quickly estimates the gain through global random sampling to broaden the search scope. These two functions synergistically guide the search process. Furthermore, we introduce an age strategy to break ties and enhance the diversity of the search path. When no improvements are made for extended iterations, a large-scale random perturbation is executed to help the algorithm escape local stagnation. Comparative experiments were conducted using 200 public hitting set instances (100 metaheuristic and 100 exact cases) from PACE-2025. The results show that, in the metaheuristic case group, DemHS improves the record on 62 instances and in the precise case group, DemHS obtains optimal solutions across all tested instances. The ablation studies demonstrate a significant positive gain from the dual-evaluation neighborhood structure on optimization results.

Vessel trajectory prediction (VTP) plays a critical role in maritime safety and intelligent navigation. Existing methods struggle to simultaneously capture long-term dependencies and nonlinear dynamic patterns in vessel movements. To address this challenge, we propose MKAIS, a novel trajectory prediction model that integrates the selective state space modeling capability of Mamba with the strong nonlinear representation power of Kolmogorov–Arnold Networks (KAN). Specifically, we design a feature-separated embedding strategy for AIS inputs (longitude, latitude, speed over ground, course over ground), followed by an MKAN module that jointly models global temporal dependencies and nonlinear dynamics. Experiments on the public ct_dma dataset demonstrate that MKAIS outperforms state-of-the-art baselines (LSTM, Transformer, TrAISformer, Mamba), achieving up to 16.65% improvement in the Haversine distance over 3 h prediction horizons. These results highlight the effectiveness and robustness of MKAIS for both short-term and long-term vessel trajectory prediction.

In recent years, it has become a hot topic to combine graph neural networks with contrastive learning, a method which has been applied not only in recommendation tasks but has also achieved impressive results in many fields, such as text processing and spatio-temporal modeling. However, existing methods are still constrained by several issues: (1) Most graph learning methods do not explore the imbalance of node and edge type distribution caused by different user–item interactions. (2) The randomness of data expansion and sampling strategies in contrastive learning may lead to confusion about the importance of key items for users. To overcome the problems, in this paper, we propose an explanation-guided contrastive recommendation model based on interactive message propagation and dual-hypergraph convolution (ECR-ID). Specifically, we designed two different interactive propagation mechanisms for the user–item dual-hypergraph sets to promote comprehensive dynamic information propagation and exchange, which further mitigates the imbalance problem of hyperedges and nodes in the hypergraph convolution, as well as the propagation loss of synergistic information between nodes. In addition, we developed an explanation-guided contrastive learning framework, which highlights the important items in user–item interactions through an explanation-based approach and guided the training of the contrastive learning framework based on the differences in the importance scores of the items, thus generating accurate positive and negative views and improving the contrastive learning performance. Finally, we integrated the contrastive learning framework with the dual-hypergraph networks based on joint training to further improve the recommendation performance. Extensive experimental evaluations on real datasets show that ECR-ID outperforms state-of-the-art recommendation algorithms. In the future, we will conduct in-depth tests based on a wider range of real-world datasets to alleviate the limitation that the existing experimental datasets all comprise data from single business services like Alibaba and Amazon, thus validating the effectiveness of our model more comprehensively.

Vessel trajectory prediction is important in maritime traffic safety and emergency management. Vessel trajectory prediction using vessel automatic identification system (AIS) data has attracted wide attention. Deep learning techniques have been widely applied to vessel trajectory prediction tasks due to their advantages in fine-grained feature learning and time series modelling. However, most deep learning-based methods use a unified approach for modelling AIS data, ignoring the diversity of AIS data and the impact of noise on prediction performance due to environmental factors. To address this issue, this study introduces a method consisting of temporal convolutional network (TCN), convolutional neural network (CNN) and convolutional long short-term memory (ConvLSTM) to predict vessel trajectories, called TCC. The model employs TCN to capture the complex correlation of the time series, utilises CNN to capture the fine-grained covariate features and then captures the dynamics and complexity of the trajectory sequences through ConvLSTM to predict vessel trajectories. Experiments are conducted on real public datasets, and the results show that the TCC model proposed in this paper outperforms the existing baseline algorithms with high accuracy and robustness in vessel trajectory prediction.