Graph Federated Learning (GFL) leverages distributed collaborative training for global graph neural networks and it is widely applied in many fields, such as social network analysis, recommendation systems and so on. However, GFL often suffers from heterogeneous label noise interferences across clients, and existing studies mainly handle label noise issues for images, where the data is independent and the labeled samples are sufficient. However, these methods are not suitable for GFL scenarios and they fail to cope with the challenges of label noise from graph-structured data, in which nodes are dependent and label information is highly sparse. To address the above problems, we propose a Two-stage Dual-branch Robust Graph Federated Learning Framework (TD-RGFL). Specifically, based on the established neighborhood conflict estimation scheme, we first design an available noisy clients identification algorithm, to effectively utilize the limited label information in graph data and capture the characteristics of noisy information. In addition, we propose a feasible dual-branch collaborative training method by means of debiased knowledge distillation, so as to fully exploit the useful data information of noisy clients and enhance the overall robustness of federated learning. Moreover, we devise a flexible federated aggregation and local model update strategy based on the proximal regularization principle, to further reduce the negative impact of noisy clients and improve the generalization ability of the global model across different clients. Finally, we implement a series of experiments, and the corresponding results indicate that our algorithm significantly outperforms existing methods under different conditions.

Domain adaptation in point cloud data is essential for improving downstream tasks in autonomous driving, robotics, and 3D modeling. 3D Foundation models, driven by scaling laws, have significantly advanced point cloud applications by embedding rich semantic knowledge of geometric structures. However, a significant gap remains between their broad zero-shot generalization capabilities and the specialized requirements of domain adaptation tasks. Furthermore, pre-training can induce a model bias towards samples that resemble the pre-training dataset. To bridge this gap, we propose an Evolving Alignment strategy to apply large-scale 3D foundation models to domain adaptation in 3D point clouds, named EvoAlign3D. Specifically, we implement a joint domain alignment strategy to align the foundation model’s feature space with a transferable feature space across the source and target domains. Meanwhile, we propose a debiased prototype evolving method, which refines class-level prototypes and optimizes pseudo-label consistency, progressively mitigating model biases and enhancing the transferability of discriminative features for better cross-domain generalization. Our method significantly improves domain adaptation classification performance on the PointDA dataset, which spans both synthetic and real-world data domains. All the code and pre-trained weights will be publicly available.

Specific emitter identification (SEI) is capable of identifying individuals from varying radiation sources, and it has been widely used in both military and civilian fields. Current deep learning-based SEI methods typically demand substantial training samples. However, their performance deteriorates significantly when confronted with abnormal or poisoned samples, such as those subjected to impersonation attacks or label flipping attacks. In this paper, we propose an efficient machine unlearning-based algorithm with knowledge distillation and noise generation, to deal with the negative effects of forgettable poisoned samples in trained SEI deep network models while keeping the identification accuracy for the retaining sample set. Specifically, we first establish a feasible two teachers-one student based knowledge distillation framework without any training restrictions, to achieve the coarse-grained student network for unlearning ability in poisoned SEI model. In addition, based on the constructed loss-maximizing noise generation model for forgettable samples, we design an available machine unlearning algorithm with impair-repair paradigm based weight manipulations, to further remove the residual poisoned sample effects in the trained student network model and improve the SEI accuracy for the retaining samples. Finally, a series of experiments are implemented on our synthetic dataset and the public ORACLE dataset, and the results demonstrate that the proposed method can achieve the average accuracy of 98.34% on the retain set and 0.54% on the forgettable set. and the efficiency of our unlearning method is almost 10 times faster than that of other retrain-based methods.

In Specific Emitter Identification (SEI), transmitters are typically distinguished through Radio Frequency Fingerprint (RFF) features. However, modulation schemes can be deliberately coupled to confound RFF information. This paper addresses modulation variation as a Domain Adaptation (DA) problem and proposes an SEI framework based on Margin Disparity Discrepancy (MDD) to enhance robustness in modulation-varying scenarios. Specifically, we first establish a theoretical tight upper bound for the discrepancy between modulation domains using MDD theory. Then, we design an adversarial network to align variable features to shorten the discrepancy between modulations. Finally, we experimented with complex modulated signals including digital and analog modulation. Numerical results indicate that our approach achieves an average improvement of over 20% in accuracy compared to classical SEI methods and outperforms traditional DA techniques.

Specific Emitter Identification (SEI) is an emitter recognition technology based on the Radio Frequency Fingerprint (RFF) of hardware. The emergence of unknown emitters is frequent in non-cooperative environments, and Open-set Specific Emitter Identification (OSSEI) based studies are becoming increasingly critical. Besides, Radio Frequency (RF) signals contain a substantial number of features that are irrelevant to emitter categories, as a result the distributions of signals are extremely sparse in the feature space. Existing OSSEI methods cannot be capable in extracting categorical representations for such signals, which may lead to wrong recognition. In this work, based on a set of Variational Auto-Encoder (VAE) models, we propose a robust OSSEI framework designed to handle class-irrelevant features. Specifically, we first use the proposed class-independent VAEs to construct categorical representations for each emitter, leveraging signal distributions in the feature space. In addition, to enhance the distinction among inter-class representations and constrain intra-class distributions, we design a supervised contrastive learning (SupCL) based method that generates positive samples for data augmentation by means of sampling from the corresponding distributions. Furthermore, we calculate the category affiliation of signals by integrating reconstruction probabilities and statistical representation features, facilitating the identification of both known and unknown emitters. Finally, we validate the effectiveness of our method from both theoretical and experimental perspectives, achieving state-of-the-art (SOTA) performance.

Traffic prediction is an indispensable component of urban planning and traffic management. Achieving accurate traffic prediction hinges on the ability to capture the potential spatio-temporal relationships among road sensors. However, the majority of existing works focus on local short-term spatiotemporal correlations, failing to fully consider the interactions of different sensors in the long-term state. In addition, these works do not analyze the influences of anomalous factors, or have insufficient ability to extract personalized features of anomalous factors, which make them ineffectively capture their spatiotemporal influences on traffic prediction. To address the aforementioned issues, We propose a global spatio-temporal fusionbased traffic prediction algorithm that incorporates anomaly awareness. Initially, based on the designed anomaly detection network, we construct an efficient anomalous factors impacting module (AFIM), to evaluate the spatio-temporal impact of unexpected external events on traffic prediction. Furthermore, we propose a multi-scale spatio-temporal feature fusion module (MTSFFL) based on the transformer architecture, to obtain all possible both long and short term correlations among different sensors in a wide-area traffic environment for accurate prediction of traffic flow. Finally, experiments are implemented based on real-scenario public transportation datasets (PEMS04 and PEMS08) to demonstrate that our approach can achieve stateof-the-art performance.

Automatic Modulation Classification (AMC) plays a critical role in cognitive radio services, and latest AMC algorithms still suffers from several challenges, such as low detection accuracy, weak robustness, high dependence on labeled samples and so on in low signal-to-noise environments. To address these problems, we propose an efficient automatic modulation classification algorithm using data and knowledge dual-driven mechanism (TDDC). Initially, we design a vision transformer (ViT)-based model to extract attribute-level knowledge from the raw signal. Besides, we propose a data-driven multi-scale feature extraction model based on skip connection, to effectively capture spatio-temporal correlations of signals.Finally, to validate the effectiveness of our approach, we conduct extensive simulation experiments using the RML2016.10a and RML2016.10b datasets.

Traditional vehicle trajectory prediction models widely exist the generalization problem towards unknown scenarios. In this paper, we address the generalization via the following ways. A conditional variational autoencoder based on invariance penalty is adopted to predict trajectory endpoints. In addition, we propose a domain division method to enhance the performance of the invariance principle and design the maneuver-related subtasks to reconstruct the consistent features of trajectories. The experiment is carried out on the INTERACTION dataset, which is well employed in the study of trajectory domains. Compared to the SOTA models, the mADE at 3s decreases from 1.16 to 0.53. The ablation study demonstrates the effectiveness of each module in our model. The results show that our method achieves excellence performance when generalized to unknown domains.

Traffic prediction is an important component in the development of Intelligent Transportation Systems (ITS). To improve the accuracy of traffic prediction, many existing studies focus on combining recurrent neural networks and graph neural networks and achieve certain effects, but these works still suffer from some limitations on realizing rigorous long-term traffic foreseeing in highly dynamic spatio-temporal environments. To address this issue, we propose an efficient skip link-based traffic prediction algorithm with multiscale feature extraction (SKMSGCN). Specifically, we firstly design an available Spatio-temporal Traffic Feature Extraction module (STFE) based on the established information fusion scheme, to sufficiently obtain long-term traffic feature correlations. In addition, by means of memory networks, we propose a Critical traffic Feature Classification module (CFC), to obtain notable features from traffic sensor nodes with different attributions and further increase traffic prediction accuracy. Moreover, in order to efficiently process the input data of enhanced traffic features from CFC module and achieve the precise traffic foreseeing, we construct a long-term traffic prediction module based on graph convolutional networks and gate recurrent units. The extensive experimental results on two typical benchmark datasets firmly demonstrate the effectiveness of the proposed SKMSGCN in quantitative aspects.

Prediction of the likely evolution of traffic scenes is a challenging task because of high uncertainties from sensing technology and the dynamic environment. It leads to failure of motion planning for intelligent agents like autonomous vehicles. In this paper, we propose a fluid-inspired model to estimate collision risk in road scenes. Multi-object states are detected and tracked, and then a stable fluid model is adopted to construct the risk field. Objects’ state spaces are used as the boundary conditions in the simulation of advection and diffusion processes. We have evaluated our approach on the public KITTI dataset; our model can provide predictions in the cases of misdetection and tracking error caused by occlusion. It proves a promising approach for collision risk assessment in road scenes.

Sensing and presenting on-road information of moving vehicles is essential for fully and semi-automated driving. It is challenging to track vehicles from affordable on-board cameras in crowded scenes. The mismatch or missing data are unavoidable and it is ineffective to directly present uncertain cues to support the decision-making. In this paper, we propose a physical model based on incompressible fluid dynamics to represent the vehicle’s motion, which provides hints of possible collision as a continuous scalar riskmap. We estimate the position and velocity of other vehicles from a monocular on-board camera located in front of the ego-vehicle. The noisy trajectories are then modeled as the boundary conditions in the simulation of advection and diffusion process. We then interactively display the animating distribution of substances, and show that the continuous scalar riskmap well matches the perception of vehicles even in presence of the tracking failures. We test our method on real-world scenes and discuss about its application for driving assistance and autonomous vehicle in the future.

Vehicle safety is the study and practice for minimizing the occurrences and consequences of traffic accidents. It is found that driver behaviors such as drowsiness, impaired driving and distraction are contributing factors to traffic accidents. In complex road surroundings, comprehensive analysis is more robust than separate evaluations which are broadly proceeded with. In this paper, we propose a vision-based nonintrusive system involving lane and driver's eye features to analyze driver behaviors. In the framework of evidence theory, evaluations of driver drowsiness and distracted and impaired driving performance are integrated to evaluate vehicle safety in real time. The system was validated in real world scenarios, and experimental results demonstrate that it is promising to improve the robustness and temporal response of vehicle safety vigilance.

Driver drowsiness influences critically the driving safety and the lack of discerning the drowsy level precisely causes failure to take measures to prevent the accidents. In this paper, a novel intelligent surveillance system is proposed to estimate driver drowsiness based on the Observer Rating of Drowsiness (ORD) model integrated into evidence theory via fusion of lane and eye features. ORD is a subjective assessment of drowsiness that is reflected in people's physical appearance, behaviors and mannerisms. Its drowsiness model in five levels, which acts as the framework in evidence theory, is used to describe the driver's state. Based on expert knowledge and data statistics, various visual eye features are studied to enhance the robustness of this system. The system is validated in real world scenarios, and experiment results demonstrate that it is promising to improve the robustness and temporal response of driver surveillance in real-time.

In this paper, driving drowsiness detection based on visual features offers a noninvasive solution to detect the driver's state. Fusion with lane and driver features is addressed in order to complement each other once any visual signs failed. Given uncertainty exists greatly, Dempster-Shafer theory is used to improve the accuracy of detection while reliability is given to present the data's robustness. Experimental results demonstrate that the performance of driving drowsiness vigilance is enhanced in the proposed framework and efficiently tolerates the failure of feature collection.