Graph Neural Networks (GNNs) are widely used in financial fraud detection due to their excellent ability on handling graph-structured financial data and modeling multilayer connections by aggregating information of neighbors. However, these GNN-based methods focus on extracting neighbor-level information but neglect a global perspective. This paper presents the concept and calculation formula of Global Confidence Degree (GCD) and thus designs GCD-based GNN (GCD-GNN) that can address the challenges of camouflage in fraudulent activities and thus can capture more global information. To obtain a precise GCD for each node, we use a multilayer perceptron to transform features and then the new features and the corresponding prototype are used to eliminate unnecessary information. The GCD of a node evaluates the typicality of the node and thus we can leverage GCD to generate attention values for message aggregation. This process is carried out through both the original GCD and its inverse, allowing us to capture both the typical neighbors with high GCD and the atypical ones with low GCD. Extensive experiments on two public datasets demonstrate that GCD-GNN outperforms state-of-the-art baselines, highlighting the effectiveness of GCD. We also design a lightweight GCD-GNN (GCD-GNNlight ) that also outperforms the baselines but is slightly weaker than GCD-GNN on fraud detection performance. However, GCD-GNNlight obviously outperforms GCD-GNN on convergence and inference speed.

LiDAR-based place recognition (LPR) technology processes 3D LiDAR point clouds and encodes them into feature descriptors, enabling mobile robots to recognize previously visited locations. This capability supports critical tasks such as loop closure detection and re-localization. With the rapid advancements in deep learning, deep learning-based LiDAR place recognition (DL-LPR) has emerged as the dominant research direction in this field. However, existing reviews on DL-LPR remain limited in scope. To address this gap, this paper focuses on DL-LPR, introducing its core concepts, system structures, and applications. It presents a coarse-to-fine classification framework to systematically categorize and review existing methods, based on two dimensions: input data structure and model architecture. Furthermore, this paper summarizes commonly used datasets and performance evaluation metrics, along with performance comparisons of representative methods. Finally, it provides an in-depth analysis of the challenges faced by DL-LPR in complex environments, such as long-term, large-scale, and dynamic settings, and offers insights into future development trends.

Deep Reinforcement Learning (DRL) is capable of learning a policy with great scene adaptation ability through interactions with the environment, and has application potential in the field of autonomous driving. However, using DRL to directly control the vehicle motion command is easy to lead to fluctuation and non-smoothness. Discrete DRL decision-making method can generate stable behavior but lose some maneuverability. Thus, the trade-off between flexibility and stability to enhance the performance of DRL policy is an important issue. In this paper, a Deep Reinforcement Learning Decision-Making algorithm based on Motion Primitives Library (MPL) in hierarchical action space is proposed to provide flexible and reliable maneuvers for autonomous driving. The upper action space contains discrete lane-changing targets, and the lower action space is mapped as a motion primitives library. In addition, model predictive control method (MPC) based on the vehicle kinematics model is used to optimize the motion primitives instantly. The performance of the proposed method is evaluated through highway simulation. The results show that the method can make the autonomous driving lane changing process safer, more efficient and more comfortable.

This paper presents a novel dynamic lane-changing trajectory planning (DLCTP) model for autonomous vehicle (AV) running in the mixed traffic environment. The proposed model fully considers the dynamics of surrounding human-driven vehicles and can work on both straight and curved roads. The first step of the DLCTP model is to decide when and where to make the lane change based on the car-following model and safety constraints. Upon decision-making, an optimal lane-changing trajectory that accounts for safety, comfort, and efficiency is generated at each time step until the lane-changing procedure is completed. CarSim-Simulink based simulation platform and three typical traffic scenarios are applied to validate the proposed DLCTP model. Experimental results show that the proposed DLCTP model can generate smooth, safe, and comfort trajectories even in complex traffic situations. The proposed DLCTP model can be employed directly on real AVs because it is easy to implement and can adapt to complex traffic environments.