Pub Date : 2025-04-14DOI: 10.1109/TETCI.2025.3551950
Faizanuddin Ansari;Abhranta Panigrahi;Swagatam Das
This study addresses the challenges of learning from long-tailed class imbalances in deep neural networks, particularly for image recognition. Long-tailed class imbalances occur when a dataset's class distribution is highly skewed, with a few head classes containing many instances and numerous tail classes having fewer instances. This imbalance becomes problematic when traditional classification methods, especially deep learning models, prioritize accuracy in the more frequent classes, neglecting the less common ones. Furthermore, these methods struggle to maintain consistent boundary fidelity—decision boundaries that are sharp enough to distinguish classes yet smooth enough to generalize well. Hard boundaries, often caused by overfitting tail classes, amplify intra-class variations, while overly soft boundaries blur distinctions between classes, reducing classification accuracy. We propose a dual-branch network with a shared feature extractor to overcome these challenges. This network uses instance and median samplers for head and medium classes and a reverse sampler for tail classes. Additionally, we implement a specialized loss function as a feature regularizer to reduce the model's sensitivity to irrelevant intra-class variations during classification. This loss function dynamically modulates feature representation alignment, promoting cohesive intra-class structures and clear inter-class separations. To achieve this, our framework incorporates two key components: Dual-Branch Sampler-Guided Mixup (DBSGM) and Adaptive Class-Aware Feature Regularizer (ACFR), which work together to balance class representation and refine decision boundaries. Integrating DBSGM and ACFR during training helps shape decision boundaries that align with class semantics. To ensure class boundaries are appropriately defined, we propose the temperature-adaptive supervised contrastive loss (TASCL) within the ACFR module, achieving the right balance between smoothness and sharpness. Our single-stage, end-to-end framework demonstrates significant performance improvements, offering a promising solution to the challenges of long-tailed class imbalances in deep learning.
{"title":"The Goldilocks Principle: Achieving Just Right Boundary Fidelity for Long-Tailed Classification","authors":"Faizanuddin Ansari;Abhranta Panigrahi;Swagatam Das","doi":"10.1109/TETCI.2025.3551950","DOIUrl":"https://doi.org/10.1109/TETCI.2025.3551950","url":null,"abstract":"This study addresses the challenges of learning from long-tailed class imbalances in deep neural networks, particularly for image recognition. Long-tailed class imbalances occur when a dataset's class distribution is highly skewed, with a few head classes containing many instances and numerous tail classes having fewer instances. This imbalance becomes problematic when traditional classification methods, especially deep learning models, prioritize accuracy in the more frequent classes, neglecting the less common ones. Furthermore, these methods struggle to maintain consistent boundary fidelity—decision boundaries that are sharp enough to distinguish classes yet smooth enough to generalize well. Hard boundaries, often caused by overfitting tail classes, amplify intra-class variations, while overly soft boundaries blur distinctions between classes, reducing classification accuracy. We propose a dual-branch network with a shared feature extractor to overcome these challenges. This network uses instance and median samplers for head and medium classes and a reverse sampler for tail classes. Additionally, we implement a specialized loss function as a feature regularizer to reduce the model's sensitivity to irrelevant intra-class variations during classification. This loss function dynamically modulates feature representation alignment, promoting cohesive intra-class structures and clear inter-class separations. To achieve this, our framework incorporates two key components: Dual-Branch Sampler-Guided Mixup (DBSGM) and Adaptive Class-Aware Feature Regularizer (ACFR), which work together to balance class representation and refine decision boundaries. Integrating DBSGM and ACFR during training helps shape decision boundaries that align with class semantics. To ensure class boundaries are appropriately defined, we propose the temperature-adaptive supervised contrastive loss (TASCL) within the ACFR module, achieving the right balance between smoothness and sharpness. Our single-stage, end-to-end framework demonstrates significant performance improvements, offering a promising solution to the challenges of long-tailed class imbalances in deep learning.","PeriodicalId":13135,"journal":{"name":"IEEE Transactions on Emerging Topics in Computational Intelligence","volume":"9 5","pages":"3650-3664"},"PeriodicalIF":5.3,"publicationDate":"2025-04-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145128547","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}
The increasing variety of energy conversion units and storage equipment connected to the multi-energy system, along with the uncertain factors posed by distributed wind and photovoltaic power generation, have made the energy flow structure of the system more complex. This complexity has created significant challenges for the frequency regulation of traditional energy hub systems. One of the characteristics of a microgrid (MG) is the use of combined heat and power (CHP) systems to generate both electrical and thermal energy at the same time. This can boost the system's dependability, efficiency, and economic performance. As a CHP's ramping capability makes it a useful tool for monitoring and controlling the MG's frequency, it will be employed in this research to achieve this goal. The complexity of the system's dynamics and set tasks throughout the course of the performance period necessitates advanced control structures for the MG with CHP systems. To address the challenges of controlling in MG with CHP systems, this research introduces a novel control structure based on deep reinforcement learning and single input interval type-2 fuzzy fractional-order proportional integral (SIT2-FFOPI) for this system. The SIT2-FFOPI serves as the main controller, with its fundamental parameters established through the utilization of the Improved Salp Swarm Algorithm (ISSA) optimization technique. An adaptive deep deterministic policy gradient (DDPG)-based actor-critic system has been developed to enhance the main controller's learning potential, thereby enabling it to more effectively address control challenges in the isolated MG. The efficacy of the suggested approach in real-time was evaluated through simulations carried out utilizing an OPAL-RT-based Hardware-in-the-Loop (HiL) configuration. As a result of this study, it was determined that the proposed controller for load disturbance, renewable energy sources (RES) power changes, and contingency circumstances in MG outperforms other controllers in terms of performance.
{"title":"Smart Energy Hub Frequency Control-Based Machine Learning","authors":"Burak Yildirim;Meysam Gheisarnejad;Mohammad Hassan Khooban","doi":"10.1109/TETCI.2025.3551991","DOIUrl":"https://doi.org/10.1109/TETCI.2025.3551991","url":null,"abstract":"The increasing variety of energy conversion units and storage equipment connected to the multi-energy system, along with the uncertain factors posed by distributed wind and photovoltaic power generation, have made the energy flow structure of the system more complex. This complexity has created significant challenges for the frequency regulation of traditional energy hub systems. One of the characteristics of a microgrid (MG) is the use of combined heat and power (CHP) systems to generate both electrical and thermal energy at the same time. This can boost the system's dependability, efficiency, and economic performance. As a CHP's ramping capability makes it a useful tool for monitoring and controlling the MG's frequency, it will be employed in this research to achieve this goal. The complexity of the system's dynamics and set tasks throughout the course of the performance period necessitates advanced control structures for the MG with CHP systems. To address the challenges of controlling in MG with CHP systems, this research introduces a novel control structure based on deep reinforcement learning and single input interval type-2 fuzzy fractional-order proportional integral (SIT2-FFOPI) for this system. The SIT2-FFOPI serves as the main controller, with its fundamental parameters established through the utilization of the Improved Salp Swarm Algorithm (ISSA) optimization technique. An adaptive deep deterministic policy gradient (DDPG)-based actor-critic system has been developed to enhance the main controller's learning potential, thereby enabling it to more effectively address control challenges in the isolated MG. The efficacy of the suggested approach in real-time was evaluated through simulations carried out utilizing an OPAL-RT-based Hardware-in-the-Loop (HiL) configuration. As a result of this study, it was determined that the proposed controller for load disturbance, renewable energy sources (RES) power changes, and contingency circumstances in MG outperforms other controllers in terms of performance.","PeriodicalId":13135,"journal":{"name":"IEEE Transactions on Emerging Topics in Computational Intelligence","volume":"9 5","pages":"3638-3649"},"PeriodicalIF":5.3,"publicationDate":"2025-04-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145128551","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}
Pub Date : 2025-04-09DOI: 10.1109/TETCI.2025.3543389
Xianghua Li;Jiao Qiao;Shu Yin;Lianwei Wu;Chao Gao;Zhen Wang;Xuelong Li
The growth of social media platforms has made it easier for fake news to spread, which poses a significant threat to authoritative news outlets, politics, and public health. Manual verification of the massive amount of online information has proven to be a daunting task, which has led to the growing interest in automatic fake news detection. Some methods that rely on news text, images, external knowledge, social contexts, or propagation graphs have demonstrated good performance. In contrast to earlier studies that focused solely on the unimodal news textual information, recent works have integrated multimodal features from various granularities, such as words, visual semantic regions, and multimodal entities, to more effectively leverage news content and align with human reading habits. However, a comprehensive review of Multimodal Fake News Detection (MFND) is still lacking, prompting our aim to complement this topic. Specifically, we present a systematic taxonomy from the perspective of cross-modal interactions. We categorize existing methods into the data-based, entity-based, and knowledge-based approaches. Connections between various works are detailed when outlining representative papers. Additionally, we introduce prevalent multimodal learning methods, present accessible MFND datasets and evaluation metrics, and analyze current research results. Finally, the promising future research directions are discussed.
{"title":"A Survey of Multimodal Fake News Detection: A Cross-Modal Interaction Perspective","authors":"Xianghua Li;Jiao Qiao;Shu Yin;Lianwei Wu;Chao Gao;Zhen Wang;Xuelong Li","doi":"10.1109/TETCI.2025.3543389","DOIUrl":"https://doi.org/10.1109/TETCI.2025.3543389","url":null,"abstract":"The growth of social media platforms has made it easier for fake news to spread, which poses a significant threat to authoritative news outlets, politics, and public health. Manual verification of the massive amount of online information has proven to be a daunting task, which has led to the growing interest in automatic fake news detection. Some methods that rely on news text, images, external knowledge, social contexts, or propagation graphs have demonstrated good performance. In contrast to earlier studies that focused solely on the unimodal news textual information, recent works have integrated multimodal features from various granularities, such as words, visual semantic regions, and multimodal entities, to more effectively leverage news content and align with human reading habits. However, a comprehensive review of Multimodal Fake News Detection (MFND) is still lacking, prompting our aim to complement this topic. Specifically, we present a systematic taxonomy from the perspective of cross-modal interactions. We categorize existing methods into the data-based, entity-based, and knowledge-based approaches. Connections between various works are detailed when outlining representative papers. Additionally, we introduce prevalent multimodal learning methods, present accessible MFND datasets and evaluation metrics, and analyze current research results. Finally, the promising future research directions are discussed.","PeriodicalId":13135,"journal":{"name":"IEEE Transactions on Emerging Topics in Computational Intelligence","volume":"9 4","pages":"2658-2675"},"PeriodicalIF":5.3,"publicationDate":"2025-04-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144687742","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}
Pub Date : 2025-03-31DOI: 10.1109/TETCI.2025.3547611
Qingling Zhu;Yeming Yang;Songbai Liu;Qiuzhen Lin;Kay Chen Tan
The evolutionary neural architecture search for generative adversarial networks (GANs) has demonstrated promising performance for generating high-quality images. However, two challenges persist, including the long search times and unstable search results. To alleviate these problems, this paper proposes a sampling and clustering-based neural architecture search algorithm for GANs, named SCGAN, which can significantly improve searching efficiency and enhance generation quality. Two improved strategies are proposed in SCGAN. First, a constraint sampling strategy is designed to limit the parameter capacity of architectures, which calculates their architecture size and discards those exceeding a reasonable parameter threshold. Second, a clustering selection strategy is applied in each architecture iteration, which integrates a decomposition selection mechanism and a hierarchical clustering mechanism to further improve search stability. Extensive experiments on the CIFAR-10 and STL-10 datasets demonstrated that SCGAN only requires 0.4 GPU days to find a promising GAN architecture in a vast search space including approximately 10$^{15}$ networks. Our best-found GAN outperformed those obtained by other neural architecture search methods with performance metric results (IS = 9.68$pm$ 0.06, FID = 5.54) on CIFAR-10 and (IS = 12.12$pm$ 0.13, FID = 12.54) on STL-10.
{"title":"SCGAN: Sampling and Clustering-Based Neural Architecture Search for GANs","authors":"Qingling Zhu;Yeming Yang;Songbai Liu;Qiuzhen Lin;Kay Chen Tan","doi":"10.1109/TETCI.2025.3547611","DOIUrl":"https://doi.org/10.1109/TETCI.2025.3547611","url":null,"abstract":"The evolutionary neural architecture search for generative adversarial networks (GANs) has demonstrated promising performance for generating high-quality images. However, two challenges persist, including the long search times and unstable search results. To alleviate these problems, this paper proposes a sampling and clustering-based neural architecture search algorithm for GANs, named SCGAN, which can significantly improve searching efficiency and enhance generation quality. Two improved strategies are proposed in SCGAN. First, a constraint sampling strategy is designed to limit the parameter capacity of architectures, which calculates their architecture size and discards those exceeding a reasonable parameter threshold. Second, a clustering selection strategy is applied in each architecture iteration, which integrates a decomposition selection mechanism and a hierarchical clustering mechanism to further improve search stability. Extensive experiments on the CIFAR-10 and STL-10 datasets demonstrated that SCGAN only requires 0.4 GPU days to find a promising GAN architecture in a vast search space including approximately 10<inline-formula><tex-math>$^{15}$</tex-math></inline-formula> networks. Our best-found GAN outperformed those obtained by other neural architecture search methods with performance metric results (IS = 9.68<inline-formula><tex-math>$pm$</tex-math></inline-formula> 0.06, FID = 5.54) on CIFAR-10 and (IS = 12.12<inline-formula><tex-math>$pm$</tex-math></inline-formula> 0.13, FID = 12.54) on STL-10.","PeriodicalId":13135,"journal":{"name":"IEEE Transactions on Emerging Topics in Computational Intelligence","volume":"9 5","pages":"3626-3637"},"PeriodicalIF":5.3,"publicationDate":"2025-03-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145128515","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}
Pub Date : 2025-03-31DOI: 10.1109/TETCI.2025.3548719
Jiashan Wan;Na Xia;Gongwen Li;Jingyang Li;Jinhua Wu;Xulei Pan;Mengqi Lian
In the field of multivariate time series prediction, capturing the dynamic relationships and complex cyclical patterns between sequences is key to improving prediction accuracy. To address this challenge, our paper introduces MSPredictor, a multi-scale dynamic graph neural network model, which uses Fast Fourier Transform for multi-scale decoupling in the frequency domain and employs Kolmogorov-Arnold Networks for multi-scale fusion, effectively extracting significant cyclical patterns. By decomposing the original series across different scales, MSPredictor accurately models complex cyclical patterns. To enhance the model's transparency and interpretability, we introduced the ClarityLens explanatory strategy, which employs visualization techniques to make the prediction process more transparent. Specifically, it displays the adjacency matrices learned at different scales, intuitively showing the dynamic correlations between series. We also visualized the proportion of different periods in the prediction results and the specific forecasting performance at each time scale. Extensive testing on multiple real-world datasets has demonstrated that the MSPredictor significantly outperforms existing benchmarks, validating its practicality and high transparency.
{"title":"MSPredictor: A Multi-Scale Dynamic Graph Neural Network for Multivariate Time Series Prediction","authors":"Jiashan Wan;Na Xia;Gongwen Li;Jingyang Li;Jinhua Wu;Xulei Pan;Mengqi Lian","doi":"10.1109/TETCI.2025.3548719","DOIUrl":"https://doi.org/10.1109/TETCI.2025.3548719","url":null,"abstract":"In the field of multivariate time series prediction, capturing the dynamic relationships and complex cyclical patterns between sequences is key to improving prediction accuracy. To address this challenge, our paper introduces MSPredictor, a multi-scale dynamic graph neural network model, which uses Fast Fourier Transform for multi-scale decoupling in the frequency domain and employs Kolmogorov-Arnold Networks for multi-scale fusion, effectively extracting significant cyclical patterns. By decomposing the original series across different scales, MSPredictor accurately models complex cyclical patterns. To enhance the model's transparency and interpretability, we introduced the ClarityLens explanatory strategy, which employs visualization techniques to make the prediction process more transparent. Specifically, it displays the adjacency matrices learned at different scales, intuitively showing the dynamic correlations between series. We also visualized the proportion of different periods in the prediction results and the specific forecasting performance at each time scale. Extensive testing on multiple real-world datasets has demonstrated that the MSPredictor significantly outperforms existing benchmarks, validating its practicality and high transparency.","PeriodicalId":13135,"journal":{"name":"IEEE Transactions on Emerging Topics in Computational Intelligence","volume":"9 6","pages":"4243-4257"},"PeriodicalIF":5.3,"publicationDate":"2025-03-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145584598","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}
Pub Date : 2025-03-27DOI: 10.1109/TETCI.2024.3372440
Rudolf Chow;Albert Y. S. Lam
Recent rapid advances in pre-trained language models, such as BERT and GPT, in natural language processing (NLP) have greatly improved the efficacy of text classifiers, easily surpassing human level performance in standard datasets like GLUE. However, most of these standard tasks implicitly assume a closed-world situation, where all testing data are supposed to lie in the same scope or distribution of the training data. Out-of-distribution (OOD) detection is the task of detecting when an input data point lies beyond the scope of the seen training set. This is becoming increasingly important as NLP agents, such as chatbots or virtual assistants, have been being deployed ubiquitously in our daily lives, thus attracting more attention from the research community to make it more accurate and robust at the same time. Recent work can be broadly categorized into two orthogonal approaches – data generative/augmentative methods and threshold/boundary learning. In this work, we follow the former and propose a method for the task based on prompting, which is known for its zero and few-shot capabilities. Generating synthetic outliers in terms of prompts allows the model to more efficiently learn OOD samples than the existing methods. Testing on nine different settings across three standard datasets used for OOD detection, our method with adaptive decision boundary is able to achieve competitive or superior performances compared with the current state-of-the-art in all cases. We also provide extensive analysis on each dataset as well as perform comprehensive ablation studies on each component of our model.
{"title":"Prompt-Based Out-of-Distribution Intent Detection","authors":"Rudolf Chow;Albert Y. S. Lam","doi":"10.1109/TETCI.2024.3372440","DOIUrl":"https://doi.org/10.1109/TETCI.2024.3372440","url":null,"abstract":"Recent rapid advances in pre-trained language models, such as BERT and GPT, in natural language processing (NLP) have greatly improved the efficacy of text classifiers, easily surpassing human level performance in standard datasets like GLUE. However, most of these standard tasks implicitly assume a closed-world situation, where all testing data are supposed to lie in the same scope or distribution of the training data. Out-of-distribution (OOD) detection is the task of detecting when an input data point lies beyond the scope of the seen training set. This is becoming increasingly important as NLP agents, such as chatbots or virtual assistants, have been being deployed ubiquitously in our daily lives, thus attracting more attention from the research community to make it more accurate and robust at the same time. Recent work can be broadly categorized into two orthogonal approaches – data generative/augmentative methods and threshold/boundary learning. In this work, we follow the former and propose a method for the task based on prompting, which is known for its zero and few-shot capabilities. Generating synthetic outliers in terms of prompts allows the model to more efficiently learn OOD samples than the existing methods. Testing on nine different settings across three standard datasets used for OOD detection, our method with adaptive decision boundary is able to achieve competitive or superior performances compared with the current state-of-the-art in all cases. We also provide extensive analysis on each dataset as well as perform comprehensive ablation studies on each component of our model.","PeriodicalId":13135,"journal":{"name":"IEEE Transactions on Emerging Topics in Computational Intelligence","volume":"9 3","pages":"2371-2382"},"PeriodicalIF":5.3,"publicationDate":"2025-03-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144148162","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}
Pub Date : 2025-03-27DOI: 10.1109/TETCI.2025.3568244
{"title":"IEEE Computational Intelligence Society Information","authors":"","doi":"10.1109/TETCI.2025.3568244","DOIUrl":"https://doi.org/10.1109/TETCI.2025.3568244","url":null,"abstract":"","PeriodicalId":13135,"journal":{"name":"IEEE Transactions on Emerging Topics in Computational Intelligence","volume":"9 3","pages":"C4-C4"},"PeriodicalIF":5.3,"publicationDate":"2025-03-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11016138","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144148134","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}
Pub Date : 2025-03-27DOI: 10.1109/TETCI.2025.3568240
{"title":"IEEE Transactions on Emerging Topics in Computational Intelligence Publication Information","authors":"","doi":"10.1109/TETCI.2025.3568240","DOIUrl":"https://doi.org/10.1109/TETCI.2025.3568240","url":null,"abstract":"","PeriodicalId":13135,"journal":{"name":"IEEE Transactions on Emerging Topics in Computational Intelligence","volume":"9 3","pages":"C2-C2"},"PeriodicalIF":5.3,"publicationDate":"2025-03-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11016149","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144148123","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}
Few-shot learning (FSL) aims to learn novel concepts with very limited labeled data. The popular FSL methods typically rely on metric learning to measure image similarity in a learned feature space. However, existing approaches often overlook the synergy between the similarity metric and feature representation, and fail to fully exploit the combination of global and local features for effective similarity measurement. In this work, we propose a novel FSL method, AdptGL-CA, which adaptively uses global and local features to boost the discrimination capability of similarity metric, while improving feature representation and generalization through attention mechanism and contrastive learning, respectively. Specifically, we design a learnable adaptive fusion strategy that uses global similarity to represent task-specific status to adaptively determine the fusion weight of local similarity, thus effectively fusing the dual similarities for better classification. Besides, the salient parts of features are highlighted using channel and spatial attentions to improve feature representation while adjusting the importance of local descriptors. As the input to the similarity metric, these more informative features further boost its discriminative ability. Moreover, a contrastive learning loss is introduced to overcome the potential overfit to base classes and learn more generic features. Additionally, we extend the PAC-Bayes-Bernstein bound to FSL setting, introducing a theoretically grounded measure for assessing generalization. Theoretical analysis validates the generalization improvement of AdptGL-CA. Comprehensive experiments indicate that AdptGL-CA achieves competitive performance with few extra parameters on multiple standard and fine-grained few-shot benchmarks, showing the effectiveness.
{"title":"AdptGL-CA: Adaptive Global-Local Metric Fusion With Contrastive Attention for Few-Shot Learning","authors":"Zhiying Song;Pengfei Wang;Xiaokang Wang;Nenggan Zheng","doi":"10.1109/TETCI.2025.3550529","DOIUrl":"https://doi.org/10.1109/TETCI.2025.3550529","url":null,"abstract":"Few-shot learning (FSL) aims to learn novel concepts with very limited labeled data. The popular FSL methods typically rely on metric learning to measure image similarity in a learned feature space. However, existing approaches often overlook the synergy between the similarity metric and feature representation, and fail to fully exploit the combination of global and local features for effective similarity measurement. In this work, we propose a novel FSL method, AdptGL-CA, which adaptively uses global and local features to boost the discrimination capability of similarity metric, while improving feature representation and generalization through attention mechanism and contrastive learning, respectively. Specifically, we design a learnable adaptive fusion strategy that uses global similarity to represent task-specific status to adaptively determine the fusion weight of local similarity, thus effectively fusing the dual similarities for better classification. Besides, the salient parts of features are highlighted using channel and spatial attentions to improve feature representation while adjusting the importance of local descriptors. As the input to the similarity metric, these more informative features further boost its discriminative ability. Moreover, a contrastive learning loss is introduced to overcome the potential overfit to base classes and learn more generic features. Additionally, we extend the PAC-Bayes-Bernstein bound to FSL setting, introducing a theoretically grounded measure for assessing generalization. Theoretical analysis validates the generalization improvement of AdptGL-CA. Comprehensive experiments indicate that AdptGL-CA achieves competitive performance with few extra parameters on multiple standard and fine-grained few-shot benchmarks, showing the effectiveness.","PeriodicalId":13135,"journal":{"name":"IEEE Transactions on Emerging Topics in Computational Intelligence","volume":"9 5","pages":"3598-3613"},"PeriodicalIF":5.3,"publicationDate":"2025-03-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145128457","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}
Pub Date : 2025-03-27DOI: 10.1109/TETCI.2025.3568242
{"title":"IEEE Transactions on Emerging Topics in Computational Intelligence Information for Authors","authors":"","doi":"10.1109/TETCI.2025.3568242","DOIUrl":"https://doi.org/10.1109/TETCI.2025.3568242","url":null,"abstract":"","PeriodicalId":13135,"journal":{"name":"IEEE Transactions on Emerging Topics in Computational Intelligence","volume":"9 3","pages":"C3-C3"},"PeriodicalIF":5.3,"publicationDate":"2025-03-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11016135","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144148081","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}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：