Federated learning (FL) facilitates collaborative training among multiple clients while preserving data privacy by eliminating raw data transmission. However, the inherent data heterogeneity among participants induces bias during collaborative learning, significantly degrading the performance of local models. Existing FL solutions face critical challenges in achieving efficient knowledge transmission, particularly with respect to insufficient information extraction or excessive communication costs, which result in slow convergence and inferior performance. To address these limitations, we propose a novel FL framework in a synergy of multi-level prototype-based contrastive learning (CL) and soft label generation, named FedMPS. The proposed method first constructs multi-level prototypes from different layers of the model to capture semantic information in high-level features and detailed information in low-level features. These prototypes are then utilized through CL to enhance intra-class discriminability and intra-class consistency in the feature space. In addition, a prototype-guided soft label generation module is introduced to model latent interclass relationships in the output space. Instead of exchanging model parameters, FedMPS transmits only prototypes and soft labels, effectively reducing global knowledge shift and communication costs. Extensive experimental studies on six publicly available datasets validate the effectiveness of the proposed method when compared to the current state-of-the-art FL approaches. The code is available at github.com/wenxinyang1026/FedMPS.

Yong Zhao, Kai Xu, Zhengqiu Zhu, Yue Hu, Zhiheng Zheng, Yingfeng Chen, Yatai Ji, Chen Gao, Yong Li, Jincai Huang. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.

GNNs have achieved remarkable performance across a range of tasks, but their reliability under distribution shifts remains a significant challenge. In particular, energy-based OOD detection methods—which compute energy scores from GNN logits—suffer from unstable performance due to a fundamental coupling between the norm and direction of node embeddings. Our analysis reveals that this coupling leads to systematic misclassification of high-norm OOD samples and hinders reliable ID–OOD separation. Interestingly, GNNs also exhibit a desirable inductive bias known as angular clustering, where embeddings of the same class align in direction. Motivated by these observations, we propose GeoEnergy (Geometric Logit Decoupling for Energy-Based OOD Detection), a plug-and-play framework that enforces hyperspherical logit geometry by normalizing class weights while preserving embedding norms. This decoupling yields more structured energy distributions, sharper intra-class alignment, and improved calibration. GeoEnergy can be integrated into existing energy-based GNNs without retraining or architectural modification. Extensive experiments demonstrate that GeoEnergy consistently improves OOD detection performance and confidence reliability across various benchmarks and distribution shifts.

Graph neural networks (GNNs) are effective machine learning models for many graph-related applications. Despite their empirical success, many research efforts focus on the theoretical limitations of GNNs, i.e., the GNNs expressive power. Early works in this domain mainly focus on studying the graph isomorphism recognition ability of GNNs, and recent works try to leverage the properties such as subgraph counting and connectivity learning to characterize the expressive power of GNNs, which are more practical and closer to real-world. However, no survey papers and open-source repositories comprehensively summarize and discuss models in this important direction. To fill the gap, we conduct a first survey for models for enhancing expressive power under different forms of definition. Concretely, the models are reviewed based on three categories, i.e., Graph feature enhancement, Graph topology enhancement, and GNNs architecture enhancement.

Confidence calibration in Graph Neural Networks (GNNs) aims to align a model's predicted confidence with its actual accuracy. Recent studies have indicated that GNNs exhibit an under-confidence bias, which contrasts the over-confidence bias commonly observed in deep neural networks. However, our deeper investigation into this topic reveals that not all GNNs exhibit this behavior. Upon closer examination of message passing in GNNs, we found a clear link between message aggregation and confidence levels. Specifically, GNNs with extensive message aggregation, often seen in deep architectures or when leveraging large amounts of labeled data, tend to exhibit overconfidence. This overconfidence can be attributed to factors like over-learning and over-smoothing. Conversely, GNNs with fewer layers, known for their balanced message passing and superior node representation, may exhibit under-confidence. To counter these confidence biases, we introduce the Adaptive Unified Label Smoothing (AU-LS) technique. Our experiments show that AU-LS outperforms existing methods, addressing both over and under-confidence in various GNN scenarios.

On-policy deep reinforcement learning algorithms have low data utilization and require significant experience for policy improvement.Reusing important data generated by an old policy can both increase the amount of data needed to update the current policy and enhance the ability to improve the current policy.However, the importance of experience is difficult to measure effectively and the direct use of generated data increases the variance of policy improvement.We propose a proximal policy optimization algorithm with integration prioritized trajectory replay (InPTR), a method that designs three priority metrics of max, mean and reward to measure the importance of trajectories.We incorporate the prioritized trajectory replay into the PPO algorithm (PTR-PPO) where truncated importance weight and loss function for PPO policy improvement under off-policy are introduced to overcome the high variance caused by large importance weights under multi-step experience.To trade-off the performance of three priority metrics in a task, we design the learner selection module to integrate the policy of the three priority metrics.We evaluate the performance of InPTR and PTR-PPO in a set of Atari discrete control tasks.The results show that our method outperforms the mainstream reinforcement learning algorithms PPO and ACER under the same size of experience, effectively improving sample efficiency.

Transformer-based trajectory optimization methods have demonstrated exceptional performance in offline Reinforcement Learning (offline RL). Yet, it poses challenges due to substantial parameter size and limited scalability, which is particularly critical in sequential decision-making scenarios where resources are constrained such as in robots and drones with limited computational power. Mamba, a promising new linear-time sequence model, offers performance on par with transformers while delivering substantially fewer parameters on long sequences. As it remains unclear whether Mamba is compatible with trajectory optimization, this work aims to conduct comprehensive experiments to explore the potential of Decision Mamba (dubbed DeMa) in offline RL from the aspect of data structures and essential components with the following insights: (1) Long sequences impose a significant computational burden without contributing to performance improvements since DeMa's focus on sequences diminishes approximately exponentially. Consequently, we introduce a Transformer-like DeMa as opposed to an RNN-like DeMa. (2) For the components of DeMa, we identify the hidden attention mechanism as a critical factor in its success, which can also work well with other residual structures and does not require position embedding. Extensive evaluations demonstrate that our specially designed DeMa is compatible with trajectory optimization and surpasses previous methods, outperforming Decision Transformer (DT) with higher performance while using 30\% fewer parameters in Atari, and exceeding DT with only a quarter of the parameters in MuJoCo.

The transition from cyber-physical-system-based (CPS-based) Industry 4.0 to cyber-physical-social-system-based (CPSS-based) Industry 5.0 brings new requirements and opportunities to current sensing approaches, especially in light of recent progress in large language models (LLMs) and retrieval augmented generation (RAG). Therefore, the advancement of parallel intelligence powered crowdsensing intelligence (CSI) is witnessed, which is currently advancing toward linguistic intelligence. In this article, we propose a novel sensing paradigm, namely conversational crowdsensing, for Industry 5.0 (especially for social manufacturing). It can alleviate workload and professional requirements of individuals and promote the organization and operation of diverse workforce, thereby facilitating faster response and wider popularization of crowdsensing systems. Specifically, we design the architecture of conversational crowdsensing to effectively organize three types of participants (biological, robotic, and digital) from diverse communities. Through three levels of effective conversation (i.e., interhuman, human–AI, and inter-AI), complex interactions and service functionalities of different workers can be achieved to accomplish various tasks across three sensing phases (i.e., requesting, scheduling, and executing). Moreover, we explore the foundational technologies for realizing conversational crowdsensing, encompassing LLM-based multiagent systems, scenarios engineering and conversational human–AI cooperation. Finally, we present potential applications of conversational crowdsensing and discuss its implications. We envision that conversations in natural language will become the primary communication channel during crowdsensing process, enabling richer information exchange and cooperative problem-solving among humans, robots, and AI.

With recent advances in sensing technologies, a myriad of spatio-temporal data has been generated and recorded in smart cities. Forecasting the evolution patterns of spatio-temporal data is an important yet demanding aspect of urban computing, which can enhance intelligent management decisions in various fields, including transportation, environment, climate, public safety, healthcare, and others. Traditional statistical and deep learning methods struggle to capture complex correlations in urban spatio-temporal data. To this end, Spatio-Temporal Graph Neural Networks (STGNN) have been proposed, achieving great promise in recent years. STGNNs enable the extraction of complex spatio-temporal dependencies by integrating graph neural networks (GNNs) and various temporal learning methods. In this manuscript, we provide a comprehensive survey on recent progress on STGNN technologies for predictive learning in urban computing. Firstly, we provide a brief introduction to the construction methods of spatio-temporal graph data and the prevalent deep-learning architectures used in STGNNs. We then sort out the primary application domains and specific predictive learning tasks based on existing literature. Afterward, we scrutinize the design of STGNNs and their combination with some advanced technologies in recent years. Finally, we conclude the limitations of existing research and suggest potential directions for future work.

Federated learning (FL) is an emerging distributed machine learning (ML) framework that operates under privacy and communication constraints. To mitigate the data heterogeneity underlying FL, clustered FL (CFL) was proposed to learn customized models for different client groups. However, due to the lack of effective client selection strategies, the CFL process is relatively slow, and the model performance is also limited in the presence of nonindependent and identically distributed (non-IID) client data. In this work, for the first time, we propose selecting participating clients for each cluster with active learning (AL) and call our method active client selection for CFL (ACFL). More specifically, in each ACFL round, each cluster filters out a small set of clients, which are the most informative clients according to some AL metrics [e.g., uncertainty sampling, query-by-committee (QBC), loss], and aggregates only its model updates to update the cluster-specific model. We empirically evaluate our ACFL approach on the public MNIST, CIFAR-10, and LEAF synthetic datasets with class-imbalanced settings. Compared with several FL and CFL baselines, the results reveal that ACFL can dramatically speed up the learning process while requiring less client participation and significantly improving model accuracy with a relatively low communication overhead.

Deep reinforcement learning (DRL) integrates the feature representation ability of deep learning with the decision-making ability of reinforcement learning so that it can achieve powerful end-to-end learning control capabilities. In the past decade, DRL has made substantial advances in many tasks that require perceiving high-dimensional input and making optimal or near-optimal decisions. However, there are still many challenging problems in the theory and applications of DRL, especially in learning control tasks with limited samples, sparse rewards, and multiple agents. Researchers have proposed various solutions and new theories to solve these problems and promote the development of DRL. In addition, deep learning has stimulated the further development of many subfields of reinforcement learning, such as hierarchical reinforcement learning (HRL), multiagent reinforcement learning, and imitation learning. This article gives a comprehensive overview of the fundamental theories, key algorithms, and primary research domains of DRL. In addition to value-based and policy-based DRL algorithms, the advances in maximum entropy-based DRL are summarized. The future research topics of DRL are also analyzed and discussed.

Traffic congestion event prediction is an important yet challenging task in intelligent transportation systems. Many existing works about traffic prediction integrate various temporal encoders and graph convolution networks (GCNs), called spatio-temporal graph-based neural networks, which focus on predicting dense variables such as flow, speed and demand in time snapshots, but they can hardly forecast the traffic congestion events that are sparsely distributed on the continuous time axis. In recent years, neural point process (NPP) has emerged as an appropriate framework for event prediction in continuous time scenarios. However, most conventional works about NPP cannot model the complex spatio-temporal dependencies and congestion evolution patterns. To address these limitations, we propose a spatio-temporal graph neural point process framework, named STGNPP for traffic congestion event prediction. Specifically, we first design the spatio-temporal graph learning module to fully capture the long-range spatio-temporal dependencies from the historical traffic state data along with the road network. The extracted spatio-temporal hidden representation and congestion event information are then fed into a continuous gated recurrent unit to model the congestion evolution patterns. In particular, to fully exploit the periodic information, we also improve the intensity function calculation of the point process with a periodic gated mechanism. Finally, our model simultaneously predicts the occurrence time and duration of the next congestion. Extensive experiments on two real-world datasets demonstrate that our method achieves superior performance in comparison to existing state-of-the-art approaches.

Meta learning is a promising paradigm to enable skill transfer across tasks. Most previous methods employ the empirical risk minimization principle in optimization. However, the resulting worst fast adaptation to a subset of tasks can be catastrophic in risk-sensitive scenarios. To robustify fast adaptation, this paper optimizes meta learning pipelines from a distributionally robust perspective and meta trains models with the measure of tail task risk. We take the two-stage strategy as heuristics to solve the robust meta learning problem, controlling the worst fast adaptation cases at a certain probabilistic level. Experimental results show that our simple method can improve the robustness of meta learning to task distributions and reduce the conditional expectation of the worst fast adaptation risk.

Deep models need a large number of labeled samples to be trained. Furthermore, in practical application settings where objects’ features are added or changed over time, it is difficult and expensive to get enough labeled samples in the beginning. Cognitive learning mechanism can actively raise the deep models’ proficiency online with a few training labels gradually. In this paper, inspired by human being’s cognition procedure to acquire new knowledge stage by stage, we develop a novel deep active recognition framework based on the analysis of models’ cognitive error knowledge to fine-tune the deep models online. The transformation of the cognitive errors is defined, and the corresponding knowledge is obtained to identify the models’ cognitive information. Based on the cognitive knowledge, the sensitive samples are selected to finely tune the models online. To avoid forgetting the previous learned knowledge, the selected prior training samples are used as the refreshening samples at the same time. The experiments demonstrate that the sensitive samples can benefit the target recognition and the cognitive learning mechanism can boost the deep models’ performance efficiently. The characterization of cognitive information can restrain the other samples’ disturbance to the models’ cognition effectively and the online training method can save mass of the time evidently. In conclusion, we introduce this work to provide a trial of thought about the cognitive lifelong learning used in deep learning scenarios.

In many domains of empirical sciences, discovering the causal structure within variables remains an indispensable task. Recently, to tackle unoriented edges or latent assumptions violation suffered by conventional methods, researchers formulated a reinforcement learning (RL) procedure for causal discovery and equipped a REINFORCE algorithm to search for the best rewarded directed acyclic graph. The two keys to the overall performance of the procedure are the robustness of RL methods and the efficient encoding of variables. However, on the one hand, REINFORCE is prone to local convergence and unstable performance during training. Neither trust region policy optimization, being computationally expensive, nor proximal policy optimization (PPO), suffering from aggregate constraint deviation, is a decent alternative for combinatory optimization problems with considerable individual subactions. We propose a trust region-navigated clipping policy optimization method for causal discovery that guarantees both better search efficiency and steadiness in policy optimization, in comparison with REINFORCE, PPO, and our prioritized sampling-guided REINFORCE implementation. On the other hand, to boost the efficient encoding of variables, we propose a refined graph attention encoder called SDGAT that can grasp more feature information without priori neighborhood information. With these improvements, the proposed method outperforms the former RL method in both synthetic and benchmark datasets in terms of output results and optimization robustness.

Travel time estimation (TTE) is a crucial task in intelligent transportation systems, which has been widely used in navigation and route planning. In recent years, several deep learning frameworks have been proposed to capture the dynamic features of road segments or intersections for travel time estimation. However, most existing works do not consider the joint features of the intersections and road segments. Moreover, most deep neural networks for TTE are designed based on empirical knowledge. Since the independent and joint features of intersections and road segments commonly vary with different datasets, the empirical deterministic neural architectures have limited adaptability to different scenarios. To tackle the above problems, we propose a novel automated deep learning framework, namely Automated Spatio-Temporal Dual Graph Convolutional Networks (Auto-STDGCN), for travel time estimation. Specifically, we propose to construct the node-wise graph and edge-wise graph to characterize the spatio-temporal features of intersections and road segments, respectively. In order to capture the joint spatio-temporal correlations of the dual graphs, a hierarchical neural architecture search approach is introduced, whose search space is composed of internal and external search space. In the internal search space, spatial graph convolution and temporal convolution operations are adopted to capture the respective spatio-temporal correlations of the dual graphs. Further, we design the external search space including the node-wise and edge-wise graph convolution operations from the internal architecture search to capture the interaction patterns between the intersections and road segments. We evaluate our proposed model Auto-STDGCN on three real-world datasets, which demonstrates that our model is significantly superior to the state-of-the-art methods. In addition, we also conduct case studies to visualize and explain the neural architectures learned by our model.

Accurate traffic prediction is a challenging task in intelligent transportation systems because of the complex spatio-temporal dependencies in transportation networks. Many existing works utilize sophisticated temporal modeling approaches to incorporate with graph convolution networks (GCNs) for capturing short-term and long-term spatio-temporal dependencies. However, these separated modules with complicated designs could restrict effectiveness and efficiency of spatio-temporal representation learning. Furthermore, most previous works adopt the fixed graph construction methods to characterize the global spatio-temporal relations, which limits the learning capability of the model for different time periods and even different data scenarios. To overcome these limitations, we propose an automated dilated spatio-temporal synchronous graph network, named Auto-DSTSGN for traffic prediction. Specifically, we design an automated dilated spatio-temporal synchronous graph (Auto-DSTSG) module to capture the short-term and long-term spatio-temporal correlations by stacking deeper layers with dilation factors in an increasing order. Further, we propose a graph structure search approach to automatically construct the spatio-temporal synchronous graph that can adapt to different data scenarios. Extensive experiments on four real-world datasets demonstrate that our model can achieve about 10% improvements compared with the state-of-art methods. Source codes are available athttps://github.com/jinguangyin/Auto-DSTSGN.

Human–computer interaction cognitive behavior (HCICB) modeling faces four deficiencies: 1) lack of a standard framework model; 2) large simulation error; 3) single simulation dimension; and 4) lack of a simulation software. To solve these deficiencies, we have carried out work in four aspects. First, we construct an HCICB model with the user, system device, and environment as the core elements, which provides a unified framework for the subsequent HCICB modeling in the Military Internet of Things (MIoT) command and control (C2) system. Second, we correct visual and motion parameters in the adaptive control of thought rational module of the Cogtool model by the commander in the loop (CIL) experiment. Third, we construct a mental workload (MW) prediction model of the MIoT C2 system based on improved visual auditory cognitive psychomotor, which realizes fast, high-precision, and quantitative MW prediction. It is added as a simulation dimension for the HCICB. Fourth, we develop MwCogtool, an HCICB prediction software that can rapidly simulate typical tasks at the design and usage stages of the MIoT C2 system, and also can output six parameters, including task completion time (TCT), MW, eye movement prepare time, eye movement execution time, motion time, and cognitive time in the whole process quickly and visually. In addition, we select 20 real users and 9 typical tasks of the MIoT C2 system to carry out the CIL verification experiment. Compared with Cogtool, MwCogtool reduces the maximum simulation error in TCT of the C2 system from 45.00% to 5.58%. The consistency of simulation results with real user data reaches 0.99. The results of the MW prediction model can significantly and negatively predict the change of real users’ eye movement, and can accurately predict the trend of MW change. Simultaneously, we build a fitting model between the mean MW prediction value and eye movement parameters.

We propose a novel automated deep learning framework, namely Automated Spatio-Temporal Dual Graph Convolutional Networks (Auto-STDGCN), for travel time estimation. Specifically, a hierarchical neural architecture search approach is introduced to capture the joint spatio-temporal correlations of intersections and road segments, whose search space is composed of internal and external search space. In the internal search space, spatial graph convolution and temporal convolution operations are adopted to capture the spatio-temporal correlations of the dual graphs. In the external search space, the node-wise and edge-wise graph convolution operations from the internal architecture search are built to capture the interaction patterns between the intersections and road segments. We conduct several experiments on two real-world datasets, and the results demonstrate that Auto-STDGCN is significantly superior to the state-of-art methods.

Grasping the dynamics of crime situation is a long standing but significant problem and plays an instructive role in the field of security and protection. Traditional methods approach the crime forecasting via stochastic equations based on physics or statistics, which may be interpretable but less efficient in real applications. Recently, some data-driven models, especially sequence generative networks, seem to be promising in capturing spatio-temporal dynamics with massive dataset available. In this paper, we process some regional crime dataset of recent fifteen years in the crime situation awareness graphs and learn latent representations with variational auto-encoder. And then Crime Generative Adversarial Network (Crime-GAN) is formulated as a new crime forecasting model for four types of crime, integrating sequence to sequence structure and Wasserstein adversarial loss. In comparison to other typical algorithms, such as Conv-RNN, Crime-GAN shows superior forecasting performance for multi-type crime in spatio-temporal scale.

Granular Computing (GrC) and its related granular modeling methodologies have received much attention recently to take advantage of information granules. The principle of justifiable granularity fundamentally guides allocation of information granularity along with its optimization. In the existing literature, two conflict criteria coverage and specificity are optimized by aggregating them as one indicator. In this study, we thoughtfully select evolutionary multi-objective optimization (EMO) approaches for the allocation of information granularity and pave a way to design a new EMO framework considering an analysis of the conflict information of these two objectives. To demonstrate the usefulness of EMO and our proposed algorithms, we present a series of experimental studies based on a typical example of granular fuzzy rule-based models. The experimental results indicate that EMO is more efficient to find optimal solutions of allocation of information granularity. Moreover, the proposed EMO method is more feasible to find a set of solutions which exhibits superior coverage or specificity as possible.

An organization is a collection of functional members for a specific purpose and their relationships. In reality, various tangible and intangible organizations affect all aspects of people's lives. This paper proposes F-statistics for group division to evaluate and define the difference degrees of groups. The results can be used as an indicator of the group number estimation, and to verify the effectiveness of the F-statistics based group number estimation method.

Link prediction in networks has become a growing concern among researchers. In this paper, various link prediction methods are compared for better prediction results in intercity transportation networks. For practical consideration, a new index is proposed to sort the results of different algorithms taking both predicting precision and number of existing links into consideration. According to this index, we find the best threshold to determine the existence of links with simulate anneal. Experiments show that the index performs well in intercity transportation networks.

Community structure is very important for many real-world networks. It has been shown that communities are overlapping and hierarchical. However, most previous methods, based on the graph model, can't investigate these two properties of community structure simultaneously. Moreover, in some cases the use of simple graphs does not provide a complete description of the real-world network. After introducing hyper graphs to describe real-world networks and defining hyper-edge similarity measurement, we propose a Hierarchical Clustering method based on Hyper-edge Similarity (HCHS) to simultaneously detect both the overlapping and hierarchical properties of complex community structure, as well as using the newly introduced community density to evaluate the goodness of a community. The examples of application to real-world networks give excellent results.

Military mission planning aims to coordinate a set of platforms with different capacities to accomplish a set of tasks under temporal, spatial and resource constraints. However, as military operations are intrinsically dynamic and uncertain, a solution (plan) corresponding to the deterministic circumstance is fragile due to unexpected events. To tackle this problem, this paper proposes a chance constrained programming model, which incorporates many uncertain factors, such as the durations, locations and resource requirements of tasks. The objective of mission planning is to coordinate the platforms to maximize the probability that all of tasks are completed successfully while satisfying the chance constraints. The problem is solved under a computational framework combining GA and Monte Carlo Simulation, the GA is used to solve the platform allocation and task scheduling while Monte Carlo Simulation is used to process the chance constraints. A mission instance is presented which demonstrates the usefulness of the proposed model and algorithm.

The goal of image quality assessment research is to design quantitative measurements for the evaluation of image quality such that it is consistent with subjective human evaluation. Inspired by intrinsic geometric structure of nature images and characteristic of visual perception, we propose a novel geometric structural distortion model for image quality assessment in this paper, which has relatively low computational complexity and clear physical meanings. The experimental results of LIVE image database show that the proposed method is consistent with the subjective assessment of human beings and has a good performance for all distortion types.

This paper presents the monotonic indices space (MIS) method used for the extended complex system capability indices effectiveness analysis. Based on the assumption that indices are monotonic with respect to the requirement measurements, an algorithm is proposed and applied to attain numerical approximation of monotonic indices requirement locus with hyperboxes. Two algorithms for acquiring intersection of several monotonic indices requirement loci are proposed, and two system analysis models based on MIS, the system evaluation model and the index sensitivity analysis model, are put forward. Finally, the models previously mentioned are used to analyze the capability indices effectiveness of a notional antistealth information system. The results show that the MIS method is promising.

In this paper, we improve the image compression method based on the theory of fuzzy information granulation (TFIG). Granulation structure of image is introduced, and the basic principle of complementary image compression based on TFIG is investigated. A new compression method is proposed based on such granulation structure. It has been testified with better effect than the original method based on TFIG by many experiments.

This paper presents an efficient decentralized Grid service discovery approach based on service ontology. It uses two techniques to improve efficiency. First, Grid information nodes are organized into community overlays of different service categories defined in service ontology. A distributed hash table (DHT) based upper layer network is constructed to provide efficient navigation between communities. Second, a simple and lightweight greedy search based service location (GSBSL) method is introduced to identify service providers with high QoS efficiently within communities. Simulation results show that, the efficiency is improved compared with existing decentralized Grid service discovery approaches, and the overhead is acceptable and controllable.