IEEE Transactions on Cognitive and Developmental Systems最新文献

{"title":"IEEE Transactions on Cognitive and Developmental Systems Information for Authors","authors":"","doi":"10.1109/TCDS.2026.3711882","DOIUrl":"https://doi.org/10.1109/TCDS.2026.3711882","url":null,"abstract":"","PeriodicalId":54300,"journal":{"name":"IEEE Transactions on Cognitive and Developmental Systems","volume":"18 4","pages":"C4-C4"},"PeriodicalIF":4.7,"publicationDate":"2026-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11653491","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148710260","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Low-Dimensional Matrix Discriminant Analysis of EEG Network Connectivity on Riemannian Manifolds for Emotion Recognition","authors":"Hao Fang;Mingyu Wang;Zhixing Zhang;Yueming Wang;Yuxiao Yang","doi":"10.1109/TCDS.2026.3661567","DOIUrl":"https://doi.org/10.1109/TCDS.2026.3661567","url":null,"abstract":"Emotion recognition from electroencephalogram (EEG) is key in developing affective brain–computer interfaces (BCIs). The functional interaction of brain networks encodes emotion. Using functional EEG network connectivity for emotion recognition is promising but remains difficult because of three challenges. First, EEG network connectivity is typically quantified by symmetric positive definite matrices represented on Riemannian manifolds, making existing Euclidean recognition methods not directly applicable. Second, the large EEG channel number leads to high-dimensional and noisy connectivity matrices that are difficult to classify. Third, aggregating EEG phase and spectral information on Riemannian manifolds remains elusive. Here, we develop a new Riemannian matrix discriminant analysis (RMDA) method for emotion recognition from high-dimensional EEG covariance, coherence, and cross-power spectrum density matrices. RMDA projects connectivity matrices onto a low-dimensional Riemannian manifold and uses <inline-formula><tex-math>$k$</tex-math></inline-formula>-nearest neighbor classification on the low-dimensional manifold for emotion recognition. RMDA further extends to the complex Fourier domain to incorporate phase information and uses Bayesian-based frequency fusion to aggregate spectral information. We show that RMDA achieves state-of-the-art recognition results on three representative EEG-emotion datasets. For subject-dependent emotion recognition, we achieve: 1) <inline-formula><tex-math>$94.37%$</tex-math></inline-formula> on the SEED dataset; 2) <inline-formula><tex-math>$63.83%$</tex-math></inline-formula> for valance and <inline-formula><tex-math>$64.77%$</tex-math></inline-formula> for arousal on the DEAP dataset; and 3) <inline-formula><tex-math>$65.00%$</tex-math></inline-formula> for valance and <inline-formula><tex-math>$62.88%$</tex-math></inline-formula> for arousal on the MAHNOB-HCI dataset; for subject-independent emotion recognition, we achieve: 1) <inline-formula><tex-math>$85.19%$</tex-math></inline-formula> on the SEED dataset; 2) <inline-formula><tex-math>$67.11%$</tex-math></inline-formula> for valance and <inline-formula><tex-math>$66.41%$</tex-math></inline-formula> for arousal on the DEAP dataset; and 3) <inline-formula><tex-math>$64.81%$</tex-math></inline-formula> for valance and <inline-formula><tex-math>$65.20%$</tex-math></inline-formula> for arousal on the MAHNOB-HCI dataset. Meanwhile, our RMDA is parameter-efficient compared to current methods. Last, our results reveal the importance of frontal connectivity in emotion recognition, which holds promise to effectively use brain network connectivity for emotion recognition in affective BCIs.","PeriodicalId":54300,"journal":{"name":"IEEE Transactions on Cognitive and Developmental Systems","volume":"18 4","pages":"873-887"},"PeriodicalIF":4.7,"publicationDate":"2026-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148710281","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"IEEE Computational Intelligence Society Information","authors":"","doi":"10.1109/TCDS.2026.3711884","DOIUrl":"https://doi.org/10.1109/TCDS.2026.3711884","url":null,"abstract":"","PeriodicalId":54300,"journal":{"name":"IEEE Transactions on Cognitive and Developmental Systems","volume":"18 4","pages":"C3-C3"},"PeriodicalIF":4.7,"publicationDate":"2026-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11653486","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148710196","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"HeteFedEEG: Prototype-Transfer Heterogeneous Federated Learning for EEG-Based Emotion Recognition","authors":"Huan Liu;Yuzhe Zhang;Chengxi Xie;Tianyu Lou;Wei Feng;Dalin Zhang","doi":"10.1109/TCDS.2026.3661120","DOIUrl":"https://doi.org/10.1109/TCDS.2026.3661120","url":null,"abstract":"Electroencephalography (EEG)-based emotion recognition is a promising avenue in affective computing due to its high temporal resolution, noninvasiveness, and cost-effectiveness. However, practical deployment faces significant challenges, including data privacy constraints, environmental noise, and substantial intersubject variability, which hinder model generalization when trained on isolated datasets. Federated learning (FL) has emerged as a compelling solution by enabling collaborative training across institutions without sharing raw data. Nevertheless, existing federated EEG-based emotion recognition (FedEER) approaches often suffer from two critical forms of heterogeneity: 1) <italic>subject heterogeneity</i>, where EEG signals vary significantly across individuals; and 2) <italic>model heterogeneity,</i> where clients adopt diverse model architectures and training strategies. To address these challenges, we propose HeteFedEEG, a heterogeneous FL framework based on prototype transfer. HeteFedEEG leverages information bottleneck (IB) theory to decouple EEG features into emotion-relevant prototypes and individual-specific representations. Instead of sharing model parameters, clients transmit perturbed emotion prototypes, enabling effective cross-institutional collaboration while preserving data privacy. A central server aggregates these prototypes into a unified emotional feature space, facilitating communication across heterogeneous models. Furthermore, to enhance learning stability, we introduce prototype regularization and a lifelong learning loss to mitigate catastrophic forgetting during local updates. Experiments on two public datasets, SJTU emotion EEG dataset (SEED) and DEAP, demonstrate that HeteFedEEG consistently outperforms state-of-the-art methods, achieving an average improvement of 2.7% in accuracy and 1.5% in F1 score. These results highlight the potential of prototype transfer for addressing heterogeneity in FedEER while ensuring privacy and generalizability. The source code is publicly available at <uri>https://github.com/XJTU-EEG/HeteFedEEG</uri>.","PeriodicalId":54300,"journal":{"name":"IEEE Transactions on Cognitive and Developmental Systems","volume":"18 4","pages":"1070-1082"},"PeriodicalIF":4.7,"publicationDate":"2026-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148710217","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Efficient Attention-Based Learning Network for Object-Task Affordance Graph Generation in Robotic Manipulation Cognition","authors":"Zhongli Wang;Jiake Lu;Shijie Guo","doi":"10.1109/TCDS.2025.3646331","DOIUrl":"https://doi.org/10.1109/TCDS.2025.3646331","url":null,"abstract":"Cognitive abilities in robotic manipulation are crucial for advancing robotic intelligence. In complex environments with diverse objects, a robot must understand task affordances between objects to enhance its cognitive manipulation abilities. Task affordances represent potential manipulation relationships between objects. Leveraging the representation and reasoning power of knowledge graphs, we model task affordances in the form of <inline-formula><tex-math>$lt$</tex-math></inline-formula>subject, relation, object<inline-formula><tex-math>$gt$</tex-math></inline-formula>, where subjects and objects are nodes, and relations are edges. The article proposes an attention-based object-task affordance graph (OTAG) generation network for robotic manipulation cognition. The proposed network effectively extracts subject, relation, and object features and combines them through multiscale channel attention mechanisms to identify task relations and generate OTAGs. To train the network, we construct a novel object-task affordance dataset (OTAD) that includes labeled task affordances across various scenarios. Experimental results demonstrate that our method significantly improves the accuracy of OTAGs and endnotes the robot with manipulation cognitive capabilities during task execution.","PeriodicalId":54300,"journal":{"name":"IEEE Transactions on Cognitive and Developmental Systems","volume":"18 4","pages":"904-915"},"PeriodicalIF":4.7,"publicationDate":"2026-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148710262","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Human-Like Robot Impedance Regulation Skill Learning From Human–Human Demonstrations","authors":"Chenzui Li;Xi Wu;Yiming Chen;Tao Teng;Xuefeng Zhang;Sylvain Calinon;Darwin Caldwell;Fei Chen","doi":"10.1109/TCDS.2025.3643445","DOIUrl":"https://doi.org/10.1109/TCDS.2025.3643445","url":null,"abstract":"Humans are experts in physical collaboration by leveraging cognitive abilities such as perception, reasoning, and decision-making to regulate compliance behaviors based on their partners’ states and task requirements. Equipping robots with similar cognitive-inspired collaboration skills can significantly enhance the efficiency and adaptability of human–robot collaboration (HRC). This article introduces an innovative human-inspired impedance regulation skill learning framework (HI-ImpRSL) for robotic systems to achieve leader-follower and mutual adaptation in multiple physical collaborative tasks. The proposed framework enables the robot to adapt its compliance based on human states and reference trajectories derived from human–human demonstrations. By integrating electromyography (EMG) signals and motion data, we extract endpoint impedance profiles and reference trajectories to construct a joint representation via imitation learning. An LSTM-based module then learns task-oriented impedance regulation policies, which are implemented through a whole-body impedance controller for online impedance adaptation. Experimental validation was conducted through collaborative transportation, two interactive Tai Chi pushing hands, and collaborative sawing tasks with multiple human subjects, demonstrating the ability of our framework to achieve human-like collaboration skills and the superior performance from the perspective of interactive forces compared with four other related methods.","PeriodicalId":54300,"journal":{"name":"IEEE Transactions on Cognitive and Developmental Systems","volume":"18 4","pages":"929-942"},"PeriodicalIF":4.7,"publicationDate":"2026-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148710284","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Hybrid Predictive Network-Based Multiagent Reconnaissance Game With Limited Perception","authors":"Xiangyuan Shen;Bin Hu;Zhi-Hong Guan;Dingxue Zhang;Tao Li","doi":"10.1109/TCDS.2025.3641401","DOIUrl":"https://doi.org/10.1109/TCDS.2025.3641401","url":null,"abstract":"Multiagent games in complex adversarial scenarios have gained significant attention. Existing methods face challenges in modeling multistage games in complex reconnaissance environments and addressing incomplete information issues. Inspired by the results of brain science research, this article proposes a hybrid predictive network (HPN) coupled with a reinforcement learning algorithm with perceptual and environmental constraints. An impulsive long short-term memory (ImpLSTM) network is designed to simulate the brain’s short-term memory mechanism. A hybrid opponent decision prediction network combining ImpLSTM and generative adversarial network (GAN) is established to enhance the prediction of opponent behavior under perception constraints. A reinforcement learning-based decision algorithm is developed to optimize decision strategies through various reward functions, enhancing the decision-making performance of agents in complex environments. Simulations and physical experiments demonstrate the performance and effectiveness of the proposed algorithm in complex games.","PeriodicalId":54300,"journal":{"name":"IEEE Transactions on Cognitive and Developmental Systems","volume":"18 4","pages":"858-872"},"PeriodicalIF":4.7,"publicationDate":"2026-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148710195","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Visual Reinforcement Learning With Multi-objective Representation Alignment","authors":"Yue Ma;Yuhu Cheng;Jiazhi Zhang;Xuesong Wang","doi":"10.1109/TCDS.2025.3634082","DOIUrl":"https://doi.org/10.1109/TCDS.2025.3634082","url":null,"abstract":"Visual reinforcement learning (VRL) aims to learn optimal policies directly from pixel data, which holds significant potential for applications in control systems characterized by data scarcity and complex environment. However, the representation learning capability of agent is significantly hindered by high-dimensional inputs, leading to suboptimal training efficiency and poor generalization in current VRL methods. Therefore, this article proposes a VRL with multiobjective representation alignment (MORA) comprising: 1) a hierarchical representation fusion (HRF) module fusing multiscale visual features; and 2) a complementary representation alignment (CRA) module harmonizing three self-supervised objectives: unidirectional prediction objective (UPO), representation consistency objective (RCO), and reconstruction alignment objective (RAO). This synergy stabilizes latent space topology while unifying task-relevant features, effectively mitigating representational collapse. Extensive experiments on DMControl demonstrate that MORA outperforms state-of-the-art VRL methods in sample efficiency and generalization for continuous control tasks. Comprehensive ablation studies validate individual contributions of each component and underscore the criticality of multiobjective synergistic optimization.","PeriodicalId":54300,"journal":{"name":"IEEE Transactions on Cognitive and Developmental Systems","volume":"18 4","pages":"956-967"},"PeriodicalIF":4.7,"publicationDate":"2026-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148710267","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"CSPER: Curiosity-Driven Safety-Prioritized Experience Replay for Mapless Navigation Using Deep Reinforcement Learning","authors":"Wendong Xiao;Liang Yuan;Teng Ran;Li He;Jianbo Zhang;Jianping Cui","doi":"10.1109/TCDS.2026.3658125","DOIUrl":"https://doi.org/10.1109/TCDS.2026.3658125","url":null,"abstract":"Due to the sparse extrinsic rewards and uniform sampling of experience replay, reinforcement learning approaches suffer from poor exploration and low sample efficiency when applied in complex navigation environments. Moreover, learning a safe behavior policy is still challenging for continuous control of mobile robots. In this article, we propose a curiosity-driven safety-prioritized experience replay (CSPER) that leverages curiosity and safety-prioritized sampling techniques to improve the exploration and sample efficiency without ignoring safety. The state prediction error is taken as the intrinsic reward to augment the normal external reward to further promote sufficient exploration. This often accompanies potentially unsafe actions. To this end, a safety priority criterion based on a temporal difference error is designed to measure the importance of the experience. The proposed replay technique takes both sample safety and efficiency into consideration and achieves efficient and safe navigation tasks. The safe policy is also integrated into the framework to further improve the policy’s safety and robustness. We combine the proposed method with DDPG, SAC, and TD3 and evaluate it in different complex environments. The experimental results show that our proposed method achieves a better performance compared to baseline algorithms in terms of the final performance, sample efficiency, and generalization.","PeriodicalId":54300,"journal":{"name":"IEEE Transactions on Cognitive and Developmental Systems","volume":"18 4","pages":"1028-1042"},"PeriodicalIF":4.7,"publicationDate":"2026-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148710271","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Boosting Extrapolation Capabilities in Model-Free Offline Reinforcement Learning","authors":"Ziru Wang;Wanli Jiang;Zeyang Liu;Xuguang Lan","doi":"10.1109/TCDS.2026.3674737","DOIUrl":"https://doi.org/10.1109/TCDS.2026.3674737","url":null,"abstract":"Model-free offline reinforcement learning learns policies from static, precollected datasets but often suffers from overestimating the values of out-of-distribution (OOD) data, which leads to biased value predictions and degraded performance. Existing methods mainly address this issue by penalizing OOD actions, but their extrapolation capabilities are limited by two key factors: 1) action-only penalties do not handle state-space extrapolation, causing cascading errors in unseen states; and 2) uniform penalties suppress potentially beneficial extrapolation, biasing learning toward suboptimal dataset behavior. To overcome these limitations, we introduce extrapoboost offline reinforcement learning (EBORL), a method that improves extrapolation by jointly addressing both OOD states and actions. EBORL leverages an ensemble of Q-networks, each independently minimizing temporal difference error on in-distribution data, while maximizing the diversity of input gradients across the Q-networks for OOD state-action pairs. Theoretically, we provide a decomposition of the extrapolation error and prove that EBORL achieves a smaller upper bound on extrapolation error compared to baseline methods. Empirically, EBORL demonstrates strong performance on D4RL tasks with fewer ensemble heads and lower computational cost, and real-world evaluations further highlight its practical applicability.","PeriodicalId":54300,"journal":{"name":"IEEE Transactions on Cognitive and Developmental Systems","volume":"18 4","pages":"968-982"},"PeriodicalIF":4.7,"publicationDate":"2026-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148710204","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}