In autonomous driving, vision-centric 3D object detection recognizes and localizes 3D objects from RGB images. However, due to high annotation costs and diverse outdoor scenes, training data often fails to cover all possible test scenarios, known as the out-of-distribution (OOD) issue. Training-free image editing offers a promising solution for improving model robustness by training data enhancement without any modifications to pre-trained diffusion models. Nevertheless, inversion-based methods often suffer from limited effectiveness and inherent inaccuracies, while recent rectified-flow-based approaches struggle to preserve objects with accurate 3D geometry. In this paper, we propose DriveFlow, a Rectified Flow Adaptation method for training data enhancement in autonomous driving based on pre-trained Text-to-Image flow models. Based on frequency decomposition, DriveFlow introduces two strategies to adapt noise-free editing paths derived from text-conditioned velocities. 1) High-Frequency Foreground Preservation: DriveFlow incorporates a high-frequency alignment loss for foreground to maintain precise 3D object geometry. 2) Dual-Frequency Background Optimization: DriveFlow also conducts dual-frequency optimization for background, balancing editing flexibility and semantic consistency. Comprehensive experiments validate the effectiveness and efficiency of DriveFlow, demonstrating comprehensive performance improvements on all categories across OOD scenarios.

Accurate odometry is crucial for mapping and localization of autonomous vehicles in unknown environment. Deep neural networks have shown significant promise in self-supervised odometry, enabling pose estimation from consecutive sensor inputs. However, existing self-supervised visual odometry methods face scale ambiguity issue due to the inaccuracy of relative depth estimation, and self-supervised LiDAR odometry methods suffer from sparse data and the difficulty of correspondence searching. To address these challenges, we propose Self-FO, a self-supervised multi-sensor fusion odometry framework that effectively integrates image and point cloud data and overcomes the limitations of single-sensor methods. First, we utilize the inferred depth map as the fusion medium for image and point cloud, and realize multi-layer visual-LiDAR feature extraction and fusion through a novel homogeneous and heterogeneous channel exchange strategies. Then, we introduce a feature alignment and pose refinement module to fine-tune the coarse pose. By leveraging 2D-3D correspondences and a cross-modal attention mechanism, this module guides the image and point cloud features to focus on consistent scene-level observations and aligns features adaptively, significantly improving the accuracy of correspondence searching and enhancing the robustness and consistency of multimodal feature representations. Extensive experiments on KITTI and KITTI-360 dataset demonstrate that Self-FO outperforms existing learning-based odometry methods, delivering superior performance and scalability.

In this article, we investigate the triggering behaviors of periodic event-triggered multiagent systems (MASs) under multiplicative false data injection (FDI) attacks. An abstraction-based traffic model is established to characterize all possible triggering behaviors under arbitrary initial states, including the minimum interevent time (MIET) and the transition relations among IETs. We further answer the following two questions: 1) how FDI attacks affect the MIET and 2) how to select the sampling period for the anomalous MIET detection. As a potential application scenario, a behavior-based anomaly detection algorithm is developed based on the proposed traffic model to identify anomalous triggering behaviors caused by attacks. Simulations demonstrate the effectiveness and practical application of the proposed results.

Federated learning (FL) has become a transformative paradigm for distributed machine learning over wireless networks. However, the performance of FL is hindered by the unreliable communication links between resource-constrained Internet of Things (IoT) devices and the central server. To overcome this challenge, we propose a novel framework that employs an unmanned aerial vehicle (UAV) as a mobile server to enhance the FL training process. By capitalizing on the UAV’s mobility, we establish strong line-of-sight connections with IoT devices, thereby enhancing communication reliability and capacity. To maximize training efficiency, we formulate a latency minimization problem that jointly optimizes bandwidth allocation, computing resources, transmit power for both the UAV and IoT devices, and the flight trajectory of the UAV. Subsequently, we analyze the required rounds of the IoT devices training and the UAV aggregation for FL convergence. Based on the convergence constraint, we transform the problem into three subproblems and develop an efficient alternating optimization algorithm to solve this problem. Additionally, we provide a thorough analysis of the algorithm’s convergence and computational complexity. Extensive numerical results demonstrate that the proposed algorithm-based scheme not only surpasses existing benchmark schemes in reducing latency up to 15.29%, but also achieves training efficiency that nearly matches the ideal scenario.

During the operation of a distributed parameter system (DPS), its working conditions typically undergo dynamic changes. Although online learning methods can enable models to adapt to new working conditions to some extent, they often confront the “catastrophic forgetting” problem, where the updated model forgets historical working conditions. On the other hand, only a few new samples can be collected during online operation, and the sparse samples make it difficult to establish accurate models for new working conditions. Therefore, achieving precise control under full working conditions remains a challenging problem. To address these challenges, this paper proposes an adaptive predictive control method based on continuous learning that achieves stable control under full working conditions by continuously identifying working conditions and triggering adaptive model updates in real time. Specifically, a spatial-temporal feature-based working condition identification method is first proposed to identify changes in working conditions automatically. Then, to address the challenge of limited data for model updating, a parameter transfer method is proposed. Simultaneously, to ensure that the updated model retains the ability to characterize historical working conditions, an Elastic Weight Consolidation (EWC) constraint is incorporated into the loss function, thus overcoming the catastrophic forgetting problem, and ensuring the updated model can represent both the historical and new working conditions. Finally, by incorporating this condition identification mechanism and adaptive predictive model into the model predictive control framework, continuous precise control of DPS can be achieved. To demonstrate the superiority of the proposed method, extensive experiments are designed. Experimental results show that the proposed method can accurately identify new working conditions and learn new condition models with only a few online samples while overcoming model mismatch of historical conditions, ultimately achieving high-precision control under full working conditions.Note to Practitioners—Motivated by the fact that DPS often operates in different conditions and online learning methods confront “catastrophic forgetting” and “sparse sample training” problems, this paper proposes an adaptive learning control method. The proposed method can learn new working condition features with small online samples while retaining the ability to represent historical working conditions, and thus achieves accurate control under full working conditions.

Confidence calibration in classification models is a vital technique for accurately estimating the posterior probabilities of predicted results, which is crucial for assessing the likelihood of correct decisions in real-world applications. Class imbalance data, which biases the model's learning and subsequently skews predicted posterior probabilities, makes confidence calibration more challenging. Especially for underrepresented classes, which are often more important and tend to have higher uncertainty, confidence calibration is more complex and essential. Unlike previous surveys that typically separately investigate confidence calibration or class imbalance, this article comprehensively investigates confidence calibration methods for deep learning-based classification models under class imbalance. First, the problem of confidence calibration under class imbalance data is outlined. Second, this article explores the impact of class imbalance data on confidence calibration in theory, providing some explanations for empirical findings in existing studies. Third, this article reviews 60 state-of-the-art confidence calibration methods under class imbalance data, divides these methods into six groups according to method differences, and systematically compares seven properties to evaluate their superiority. Then, some commonly used and emerging evaluation methodology are summarized, including public datasets and evaluation metrics. Subsequently, this article performs necessary comparative experiments to provide better guidelines and insights to the readership. Finally, we discuss several application fields and promising research directions that serve as a guideline for future studies.

This paper provides an account of new measures of coupling developed to assess the reusability of Java components retrieved from the internet by a search engine. These measures differ from the majority of established metrics in two respects: they reflect the degree to which entities are coupled or resemble each other, and they take account of indirect couplings or similarities. An empirical comparison of the new measures with eight established metrics is described. The new measure is shown to be consistently superior at ranking components according to their reusability.