IEEE Transactions on Emerging Topics in Computational Intelligence最新文献

{"title":"Membrane Potential-Driven Adaptive Threshold Plasticity for SNNs: A Bio-Inspired Mechanism Combining Inverse Depolarization Rate and Proportional Membrane Potential Dynamics","authors":"Na Shan;Yibin Lin;Chuxi Xie;Kunlun Wu;Weidong Ma;Chaoxu Mu;Xiaorong Yan;Anguo Zhang","doi":"10.1109/TETCI.2025.3631624","DOIUrl":"https://doi.org/10.1109/TETCI.2025.3631624","url":null,"abstract":"While spiking neural networks (SNNs) have demonstrated remarkable efficiency in neuromorphic computing by emulating biological neuronal dynamics, their learning capabilities remain constrained by predominant focus on synaptic plasticity. This limitation overlooks critical neurobiological evidence showing that intrinsic neuronal plasticity, particularly dynamic threshold adaptation, plays an essential role in balancing neural responsiveness and signal fidelity. Inspired by two neurophysiological principles governing threshold regulation: 1) the inverse correlation between spiking thresholds and preceding depolarization rates, and 2) the proportional relationship between thresholds and average membrane potentials, we propose a Membrane Potential-Driven Adaptive Threshold Plasticity (MPD-ATP) framework. This biologically grounded mechanism establishes a dual-pathway control system where instantaneous depolarization rates and sustained membrane potential states jointly modulate neuronal thresholds through an adaptive scaling factor. The instantaneous depolarization rate dynamically lowers thresholds during strong input bursts, while the sustained average membrane potential adjusts the baseline threshold to stabilize firing during sparse input. This complementary regulation improves precision and robustness. Extensive evaluations on static (CIFAR-10/100) and neuromorphic (CIFAR10-DVS, DVSGesture) benchmarks demonstrate that MPD-ATP-enhanced networks achieve superior classification accuracy with enhanced noise robustness. Systematic ablation studies reveal that the coordinated interaction between depolarization-sensitive and membrane potential-proportional threshold adjustments is critical for preventing signal saturation in high-activity networks while mitigating under-activation in sparse-input scenarios.","PeriodicalId":13135,"journal":{"name":"IEEE Transactions on Emerging Topics in Computational Intelligence","volume":"10 2","pages":"1713-1725"},"PeriodicalIF":6.0,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148513686","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":"Architecture Transferability Quantifying for Multi-Task Neural Architecture Search","authors":"Tingjie Zhang;Hai-Lin Liu;Songyi Xiao;Lei Chen","doi":"10.1109/TETCI.2025.3647652","DOIUrl":"https://doi.org/10.1109/TETCI.2025.3647652","url":null,"abstract":"Multi-task neural architecture search (MTNAS) transfers architectural knowledge on neural models between different tasks, with the basic aim of optimizing neural architectures. MTNAS enables effective architectural patterns contained in the source task to be transferred and reused in the target task, which is known as transferability. However, ranking disorder between the source and target task degrades architecture transferability. To address this issue, we devise an architecture transferability quantifying method (ATQ). First, our data-agnostic method maps neural architectures into architecture embedding vectors for subsequent transferability prediction. Second, we apply transfer rank, an instance-based classifier, to alleviate transferability degradation problem. In this work, ATQ is incorporated into an evolutionary multi-task NAS algorithm (KTNAS), to select candidate architectures with better transferability. Extensive experiments are conducted on multiple benchmark datasets, such as NASBench-201, TransNAS-Bench-101 and DARTs. Experimental results show that KTNAS outperforms peer MTNAS algorithms in search efficiency and downstream task performance. Ablation study demonstrates the vital importance of transfer rank on enhancing transfer performance.","PeriodicalId":13135,"journal":{"name":"IEEE Transactions on Emerging Topics in Computational Intelligence","volume":"10 2","pages":"1753-1764"},"PeriodicalIF":6.0,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148508992","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":"Adaptive Multi-View Fuzzy Clustering With Diverse Anchor Based Guidance","authors":"Wei Zhang;Andong Li;Te Zhang;Zhaohong Deng;Weiping Ding;Guanjin Wang","doi":"10.1109/TETCI.2025.3619520","DOIUrl":"https://doi.org/10.1109/TETCI.2025.3619520","url":null,"abstract":"Fuzzy clustering has received significant attention due to its robust capability for handling uncertain data and its strong clustering performance. Recently, multi-view fuzzy clustering methods have become a popular research focus for multi-view data. However, existing methods often suffer from low computational efficiency because of the extensive distance calculations required during iterative optimization. Moreover, the challenge of adaptively and comprehensively extracting discriminative information across views to enhance clustering performance remains largely unresolved. To address these issues, this paper proposes a new adaptive multi-view fuzzy clustering method based on diverse anchors guidance (AMVFC_DAG). Based on anchor graph theory, the proposed method first introduces a new diverse anchor graph learning method with dual information exploration to capture both multi-scale common and specific information across views. To further improve clustering efficiency, the proposed method integrates this diverse anchor graph learning into a unified multi-view clustering framework with multi-scale anchors, enabling mutual enhancement between components and thereby boosting overall performance. Furthermore, adaptive scale and view weighting are incorporated to enable flexible clustering. Finally, we propose an adaptive fusion mechanism that combines the learned anchor graphs with an anchor membership matrix, to produce the final cluster assignments. Extensive experiments on multiple multi-view datasets demonstrate the proposed method’s effectiveness in both clustering performance and computational efficiency.","PeriodicalId":13135,"journal":{"name":"IEEE Transactions on Emerging Topics in Computational Intelligence","volume":"10 2","pages":"2061-2074"},"PeriodicalIF":6.0,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148511875","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":"Brain-Inspired Broad Learning Framework: Spiking Neural Network for Enhanced EEG Prediction","authors":"Yaodong Wang;Yiping Zuo;Dan Chen;Weiping Tu;Jingying Chen;Tengfei Gao;Xiaoli Li","doi":"10.1109/TETCI.2025.3647698","DOIUrl":"https://doi.org/10.1109/TETCI.2025.3647698","url":null,"abstract":"Real-time decoding of brain signals is critical for applications ranging from Brain-Computer Interface (BCI) to clinical neurodiagnostics. While various learning methods are used for predicting EEG streams, Spiking Neural Networks (SNNs) offer a compelling approach due to their sparse, event-driven computation, which naturally aligns with the temporal dynamics of neural signals. However, applying SNNs to high-dimensional, evolving EEG data reveals critical limitations: deep architectures are prone to convergence failure from excessive signal diffusion, and adapting to EEG stream requires computationally expensive global retraining. To address these limitations, this study proposes a Brain-inspired Broad Learning framework (<inline-formula><tex-math>$B^{2},L$</tex-math></inline-formula>-<inline-formula><tex-math>$SNN$</tex-math></inline-formula>). Mimicking hippocampal synaptic mechanisms, the framework involves: 1) a dual-layer, parallel architecture with Random Vector Functional-Link (RVFL) mappings that enables robust pattern interpretation while addressing the diffusion issue; and 2) a synaptic growth-inspired optimization that achieves efficient incremental learning by dynamically extending the network, eliminating the need for iterative global updates. Experiments on five benchmark datasets demonstrate that <inline-formula><tex-math>$B^{2},L$</tex-math></inline-formula>-<inline-formula><tex-math>$SNN$</tex-math></inline-formula> significantly outperforms competing SNN variants and other methods. For instance, on a motor imagery task, it achieves 94.45% accuracy while reducing inference time by 64.9%. In dynamic tests, it maintained 92.37% accuracy with only an additional 2.74 seconds of training, proving its efficient incremental learning capability. Overall, this study provides an effective alternative SNN design, supporting the advancement of sophisticated neuroengineering applications.","PeriodicalId":13135,"journal":{"name":"IEEE Transactions on Emerging Topics in Computational Intelligence","volume":"10 2","pages":"1284-1297"},"PeriodicalIF":6.0,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148512259","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":"Topological Elastic Graph Convolutional Networks for Spatial-Temporal Sequence Forecasting","authors":"Yiwen Wang;Meiling Xu;Lixin Tang","doi":"10.1109/TETCI.2024.3449900","DOIUrl":"https://doi.org/10.1109/TETCI.2024.3449900","url":null,"abstract":"Spatial-temporal sequence prediction is a hotspot problem, which is very important to the development of society. Due to the complexity and mutual influence of different regions, spatial-temporal data has complicated and dynamic spatial-temporal correlation. How to effectively capture the underlying information from data to get better predictions becomes a challenge. Current work often involves a given graph structure to build spatial dependency, but it ignores the dynamic trend and context information of data. The potential spatiotemporal dependence obtained through the similarity of time series is easily affected by external factors and it is not robust. The graph convolutional network has been proven to be a global smoothing low-pass filter, which lacks the ability of local smoothing and obtaining high-frequency information. Therefore, we proposed spatial-temporal topological elastic graph convolutional networks (STTEGCN), firstly persistent homology is introduced to acquire the topological correlation and to acquire a topological graph that can better express the potential correlation between data. Introduce an elastic message-passing mechanism to obtain global and local adaptive smoothing ability. At the same time, the high-frequency filter is used to obtain useful high-frequency information, and then mining more efficient and valuable information expression. The gated hybrid dilated convolution is put forward to strengthen the capability of the model to obtain long and short temporal features, so as to boost the predictability of the model. We conducted experimental simulations on multiple public transportation and meteorological data. The results indicated that STTEGCN has the most advanced behavior over other baselines.","PeriodicalId":13135,"journal":{"name":"IEEE Transactions on Emerging Topics in Computational Intelligence","volume":"10 2","pages":"1229-1240"},"PeriodicalIF":6.0,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148512276","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":"MRM: Improving Multihop Reasoning Capability for LLMs With Multistage QA Retrieval","authors":"Wei Li;Yanqing Zhang;Guifang Sun;Chao Ma;Jipeng Qiang;Jiaxing Shen","doi":"10.1109/TETCI.2025.3642285","DOIUrl":"https://doi.org/10.1109/TETCI.2025.3642285","url":null,"abstract":"Large Language Models (LLMs) continue to face significant challenges in improving their reasoning capabilities, particularly for knowledge-intensive multi-hop questions. Existing approaches attempt to address this hurdle through generating reasoning procedures or retrieving additional knowledge, while encountering limitations. Conventional prompting methods often produce undesired reasoning traces, while retrieving pertinent knowledge from external data sources remains problematic. In this paper, we introduce a novel approach to enhance LLM reasoning capabilities: the Multistage Retrieval Method (<bold>MRM</b>). It comprises two key components: plan design and implementation with retrieval, utilizing four specialized stages: <bold>Plan Stage</b>, <bold>Extraction Stage</b>, <bold>Reasoning Stage</b>, and <bold>Output Stage</b>. The <bold>MRM</b> process begins by constructing a prompt with demonstration examples and an instruction, guiding to design a logically coherent multi-step plan in a standardized format in <bold>Plan Stage</b>. To acquire the necessary knowledge for reasoning, we employ an innovative question-answer strategy by generating a question at each step, followed by document retrieval using vector-based methods. The retrieved documents and the generated questions are then fed into the <bold>Extraction Stage</b> to extract relevant knowledge, which are utilized to execute the next step with <bold>Reasoning Stage</b>. It helps mitigate the risk of acquiring irrelevant information by performing retrieval prior. We repeat this process until the final step of the plan, whereupon we input all extracted knowledge and the original question into the <bold>Output Stage</b> to generate the final answer. Experimental results demonstrate the effectiveness of <bold>MRM</b> over SOTA methods, achieving the highest accuracy scores across both common-sense and symbolic reasoning tasks in four public datasets.","PeriodicalId":13135,"journal":{"name":"IEEE Transactions on Emerging Topics in Computational Intelligence","volume":"10 2","pages":"1700-1712"},"PeriodicalIF":6.0,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148512605","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":"PMGDA: A Preference-Based Multiple Gradient Descent Algorithm","authors":"Xiaoyuan Zhang;Xi Lin;Qingfu Zhang","doi":"10.1109/TETCI.2025.3526459","DOIUrl":"https://doi.org/10.1109/TETCI.2025.3526459","url":null,"abstract":"It is desirable in many multi-objective machine learning applications, such as multi-task learning with conflicting objectives, multi-objective reinforcement learning, to find a Pareto solution that can match a given preference of a decision maker. These problems are often large-scale with available gradient information but cannot be handled very well by the existing algorithms. To tackle this issue, this paper proposes a novel predict-and-correct framework for locating a Pareto solution that fits the preference of a decision maker. In the proposed framework, a constraint function is introduced in the search progress to align the solution with a user-specific preference, which can be optimized simultaneously with multiple objective functions. Experimental results show that our proposed method can efficiently find a particular Pareto solution under the demand of a decision maker for standard multiobjective benchmark, multi-task learning, multi-objective reinforcement learning problems with more than thousands of decision variables.","PeriodicalId":13135,"journal":{"name":"IEEE Transactions on Emerging Topics in Computational Intelligence","volume":"10 2","pages":"1098-1110"},"PeriodicalIF":6.0,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148510430","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":"Data Stream Driven Dynamic Multiobjective Optimization Using Surrogate Transfer","authors":"Zhening Liu;Handing Wang;Maoguo Gong;Yaochu Jin","doi":"10.1109/TETCI.2025.3526505","DOIUrl":"https://doi.org/10.1109/TETCI.2025.3526505","url":null,"abstract":"Most existing dynamic multiobjective evolutionary algorithms assume that the real objective function evaluations are easily acquired during the optimization process. However, in some situations, the optimization algorithm cannot invoke the objective functions and is exclusively driven by the solutions within the continuously incoming data stream. Although these solutions come with their own objective values, the objective function associated with the solutions in the data stream experiences continuous changes across time. These problems are called data stream driven dynamic multiobjective optimization problems (DDMOPs). The difficulties of the DDMOPs are tracking the optimal solutions based on the dynamic data streams and extracting the historical information to improve the optimization performance. In this work, a multiobjective evolutionary algorithm framework (ST-MOEA) is proposed to address the above challenges. ST-MOEA uses the Kriging and RBFN models to separately substitute the objective functions and guide the optimization process. Two effective mechanisms named the surrogate transfer and reliable solution selection strategies are also proposed to further improve the algorithm performance. The surrogate transfer strategy transfers the surrogate hyperparameters used in the historical data streams to the current surrogate models to enhance the surrogate accuracies, and the reliable solution selection strategy identifies dependable solutions from the populations generated by two surrogates in order to ultimately construct the optimal solution set. During the experiment process, the proposed framework is embedded into three different multiobjective evolutionary algorithms and compared with two surrogate-assisted evolutionary algorithms on a wide range of test problems with varying decision dimensions and different levels of severity in objective function changes. The experiment results validate the effectiveness of the two proposed strategies and demonstrate the overall advantage of ST-MOEA to the compared algorithms.","PeriodicalId":13135,"journal":{"name":"IEEE Transactions on Emerging Topics in Computational Intelligence","volume":"10 2","pages":"1084-1097"},"PeriodicalIF":6.0,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148510714","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":"FullTransNet: Full Transformer With Local-Global Attention for Video Summarization","authors":"Libin Lan;Lu Jiang;Tianshu Yu;Xiaojuan Liu;Zhongshi He","doi":"10.1109/TETCI.2025.3634718","DOIUrl":"https://doi.org/10.1109/TETCI.2025.3634718","url":null,"abstract":"Video summarization aims to generate a compact, informative, and representative synopsis of raw videos, which is crucial for browsing, analyzing, and understanding video content. Dominant approaches in video summarization primarily rely on recurrent or convolutional neural networks, and more recently on encoder-only transformer architectures.However, these methods typically suffer from several limitations in parallelism, modeling long-range dependencies, and providing explicit generative capabilities.To address these issues, we propose a transformer-like architecture named FullTransNet with two-fold ideas. First, it uses a full transformer with an encoder-decoder structure as an alternative architecture for video summarization. As the full transformer is specifically designed for sequence transduction tasks, its direct application to video summarization is both intuitive and effective. Second, it replaces the standard full attention mechanism with a combination of local and global sparse attention, enabling the model to capture long-range dependencies while significantly reducing computational costs. This local-global sparse attention is applied exclusively at the encoder side, where the majority of computations occur, further enhancing efficiency. Extensive experiments on two widely used benchmark datasets, SumMe and TVSum, demonstrate that our model achieves F-scores of 54.4% and 63.9%, respectively, while maintaining relatively low computational and memory requirements. These results surpass the second-best performing methods by 0.1% and 0.3%, respectively, verifying the effectiveness and efficiency of FullTransNet.","PeriodicalId":13135,"journal":{"name":"IEEE Transactions on Emerging Topics in Computational Intelligence","volume":"10 2","pages":"1185-1200"},"PeriodicalIF":6.0,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148511626","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":"PSINet: Parallel Space Interaction Network for Human Pose Estimation","authors":"Dong Wang;Youcheng Cai;Yiming Tang;Xiaoping Liu","doi":"10.1109/TETCI.2025.3647687","DOIUrl":"https://doi.org/10.1109/TETCI.2025.3647687","url":null,"abstract":"Human pose estimation is sensitive to the spatial resolution of feature representations in a single image. While multi-scale features are often exploited to improve keypoint localization, the effectiveness of parallel spatial learning across multiple uniform-sized spaces remains underexplored. To address this, To address this, we propose the Parallel Space Interaction Network (PSINet), which splits input features into multiple identical parallel spaces to facilitate efficient spatial feature learning for human pose estimation. Based on the multiple spaces, we refine the vanilla vision Transformer by proposing a Split-Head Transformer (SHT) to enhance global interaction within parallel spaces. Furthermore, we introduce Parallel Space Fusion instrument to strengthen the capacity of the PSINet by acquiring the relationship between multiple parallel spaces. Extensive experiments demonstrate that PSINet achieves state-of-the-art accuracy after data augmentation, running at 180 FPS on a single GTX2080Ti, offering a superior trade-off between accuracy and efficiency.","PeriodicalId":13135,"journal":{"name":"IEEE Transactions on Emerging Topics in Computational Intelligence","volume":"10 2","pages":"1892-1904"},"PeriodicalIF":6.0,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148511813","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}