Knowledge-Based Systems最新文献_第2页

Learning prototypes from background and latent objects for few-shot semantic segmentation

IF 7.2 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Knowledge-Based Systems

Pub Date : 2025-02-27 DOI: 10.1016/j.knosys.2025.113218

Yicong Wang , Rong Huang , Shubo Zhou , Xueqin Jiang , Zhijun Fang

Few-shot semantic segmentation (FSS) aims to segment target object within a given image supported by few samples with pixel-level annotations. Existing FSS framework primarily focuses on target area for learning a target-object prototype while directly neglecting non-target clues. As such, the target-object prototype has not only to segment the target object but also to filter out non-target area simultaneously, resulting in numerous false positives. In this paper, we propose a background and latent-object prototype learning network (BLPLNet), which learns prototypes from not only the target area but also the non-target counterpart. From our perspective, the non-target area is delineated into background full of repeated textures and salient objects, refer to as latent objects in this paper. Specifically, a background mining module (BMM) is developed to specially learn a background prototype by episodic learning. The learned background prototype replaces the target-object one for background filtering, reducing the false positives. Moreover, a latent object mining module (LOMM), based on self-attention mechanism, works together with the BMM for learning multiple soft-orthogonal prototypes from latent objects. Then, the learned latent-object prototypes, which condense the general knowledge of objects, are used in a target object enhancement module (TOEM) to enhance the target-object prototype with the guidance of affinity-based scores. Extensive experiments on PASCAL-5

^{i}

and COCO-20

^{i}

datasets demonstrate the superiority of the BLPLNet, which outperforms state-of-the-art methods by an average of 0.60% on PASCAL-5

^{i}

. Ablation studies validate the effectiveness of each component, and visualization results indicate that the learned latent-object prototypes indeed convey the general knowledge of objects.

{"title":"Learning prototypes from background and latent objects for few-shot semantic segmentation","authors":"Yicong Wang , Rong Huang , Shubo Zhou , Xueqin Jiang , Zhijun Fang","doi":"10.1016/j.knosys.2025.113218","DOIUrl":"10.1016/j.knosys.2025.113218","url":null,"abstract":"<div><div>Few-shot semantic segmentation (FSS) aims to segment target object within a given image supported by few samples with pixel-level annotations. Existing FSS framework primarily focuses on target area for learning a target-object prototype while directly neglecting non-target clues. As such, the target-object prototype has not only to segment the target object but also to filter out non-target area simultaneously, resulting in numerous false positives. In this paper, we propose a background and latent-object prototype learning network (BLPLNet), which learns prototypes from not only the target area but also the non-target counterpart. From our perspective, the non-target area is delineated into background full of repeated textures and salient objects, refer to as latent objects in this paper. Specifically, a background mining module (BMM) is developed to specially learn a background prototype by episodic learning. The learned background prototype replaces the target-object one for background filtering, reducing the false positives. Moreover, a latent object mining module (LOMM), based on self-attention mechanism, works together with the BMM for learning multiple soft-orthogonal prototypes from latent objects. Then, the learned latent-object prototypes, which condense the general knowledge of objects, are used in a target object enhancement module (TOEM) to enhance the target-object prototype with the guidance of affinity-based scores. Extensive experiments on PASCAL-5<span><math><msup><mrow></mrow><mrow><mi>i</mi></mrow></msup></math></span> and COCO-20<span><math><msup><mrow></mrow><mrow><mi>i</mi></mrow></msup></math></span> datasets demonstrate the superiority of the BLPLNet, which outperforms state-of-the-art methods by an average of 0.60% on PASCAL-5<span><math><msup><mrow></mrow><mrow><mi>i</mi></mrow></msup></math></span>. Ablation studies validate the effectiveness of each component, and visualization results indicate that the learned latent-object prototypes indeed convey the general knowledge of objects.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"314 ","pages":"Article 113218"},"PeriodicalIF":7.2,"publicationDate":"2025-02-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143519213","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Fine-grained local label correlation for multi-label classification

IF 7.2 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Knowledge-Based Systems

Pub Date : 2025-02-27 DOI: 10.1016/j.knosys.2025.113210

Tianna Zhao , Yuanjian Zhang , Duoqian Miao , Witold Pedrycz

Comprehensive learning label correlation is conducive to boosting the accuracy of multi-label classification. While existing methods focus on exploring the correlation-aware original features or latent subspaces, they often overlook the role of correlation in deducing local structures. The oversight can result in suboptimal topic-based label correlation estimation and thus incur information loss. In contrast to the conventional single-granularity-based learning for local label correlation, we propose a multi-granularity correlation-based feature augmentation (MGOFA) model. MGOFA consists of three components that progressively refine the granularity of label correlation: granular-based feature augmentation for relative neighborhood-based class tendency estimation, granular-based latent topic mining for tendency-aware topic modeling, and fine-grained label correlation mining for augmented local label correlation learning. The information on neighborhood-based similarity between instances is explicitly leveraged and contributes to the model two-fold. Firstly, it induces the prototypes of latent topics, which share more knowledge with the label association. Secondly, it refines the discriminative granularity of the model by integrating it with the original features. Such a formulation simulates the viewpoint of human decision-making by automatically determining optimal solutions on both data and knowledge from coarse and refined granularity, respectively. Extensive comparisons completed of ten benchmarks demonstrate that MGOFA outperforms the state-of-the-art methods with satisfying convergence and sensitivity.

{"title":"Fine-grained local label correlation for multi-label classification","authors":"Tianna Zhao , Yuanjian Zhang , Duoqian Miao , Witold Pedrycz","doi":"10.1016/j.knosys.2025.113210","DOIUrl":"10.1016/j.knosys.2025.113210","url":null,"abstract":"<div><div>Comprehensive learning label correlation is conducive to boosting the accuracy of multi-label classification. While existing methods focus on exploring the correlation-aware original features or latent subspaces, they often overlook the role of correlation in deducing local structures. The oversight can result in suboptimal topic-based label correlation estimation and thus incur information loss. In contrast to the conventional single-granularity-based learning for local label correlation, we propose a multi-granularity correlation-based feature augmentation (MGOFA) model. MGOFA consists of three components that progressively refine the granularity of label correlation: granular-based feature augmentation for relative neighborhood-based class tendency estimation, granular-based latent topic mining for tendency-aware topic modeling, and fine-grained label correlation mining for augmented local label correlation learning. The information on neighborhood-based similarity between instances is explicitly leveraged and contributes to the model two-fold. Firstly, it induces the prototypes of latent topics, which share more knowledge with the label association. Secondly, it refines the discriminative granularity of the model by integrating it with the original features. Such a formulation simulates the viewpoint of human decision-making by automatically determining optimal solutions on both data and knowledge from coarse and refined granularity, respectively. Extensive comparisons completed of ten benchmarks demonstrate that MGOFA outperforms the state-of-the-art methods with satisfying convergence and sensitivity.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"314 ","pages":"Article 113210"},"PeriodicalIF":7.2,"publicationDate":"2025-02-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143528815","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Weakly-supervised spatial–temporal video grounding via spatial–temporal annotation on a single frame

IF 7.2 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Knowledge-Based Systems

Pub Date : 2025-02-27 DOI: 10.1016/j.knosys.2025.113200

Shu Luo, Shijie Jiang, Da Cao, Huangxiao Deng, Jiawei Wang, Zheng Qin

The task of weakly-supervised spatial–temporal video grounding, where model training only relies on video-sentence pairs, has garnered considerable attention. Its objective is to identify and localize spatial–temporal regions within a video that correspond to objects or events described in a query sentence. Existing approaches frame this task as a multiple instance learning (MIL) problem, where a bag is constructed for each frame and the same sentence is assigned to all frame bags. However, this approach can lead to false-positive frames as not all frames necessarily correspond to the query sentence. Additionally, region proposals in each frame are typically generated by pre-trained object detection models, which primarily focus on core regions and may result in inaccurate object or event localization. To address these issues, we propose annotating a spatial–temporal region in a single frame, which provides a simple yet effective means to enhance grounding performance without incurring significant additional cost. Specifically, we innovatively contribute a spatial–temporal MIL framework. In the temporal-level MIL, by applying Gaussian weighting to the frames of a video, we assign higher weights to the frames that are close to the annotated frame, while lower weights are assigned to frames that are further away. In the spatial-level MIL, we propose regions in the each frame and compute their similarity with the annotated bounding box, selecting regions with higher similarity scores for training. Ultimately, temporal-level and spatial-level MILs are integrated to jointly optimize the accuracy of both types of grounding. Through experimental evaluations on two re-annotated datasets, our proposed framework has been demonstrated to exhibit superiority in terms of both overall performance comparison and detailed micro-level analyses. Compared to the latest weakly-supervised methods on the VidSTG dataset, our method improves the temporal localization performance by at least 10.35% and the spatial localization performance by at least 11.89%.

{"title":"Weakly-supervised spatial–temporal video grounding via spatial–temporal annotation on a single frame","authors":"Shu Luo, Shijie Jiang, Da Cao, Huangxiao Deng, Jiawei Wang, Zheng Qin","doi":"10.1016/j.knosys.2025.113200","DOIUrl":"10.1016/j.knosys.2025.113200","url":null,"abstract":"<div><div>The task of weakly-supervised spatial–temporal video grounding, where model training only relies on video-sentence pairs, has garnered considerable attention. Its objective is to identify and localize spatial–temporal regions within a video that correspond to objects or events described in a query sentence. Existing approaches frame this task as a multiple instance learning (MIL) problem, where a bag is constructed for each frame and the same sentence is assigned to all frame bags. However, this approach can lead to false-positive frames as not all frames necessarily correspond to the query sentence. Additionally, region proposals in each frame are typically generated by pre-trained object detection models, which primarily focus on core regions and may result in inaccurate object or event localization. To address these issues, we propose annotating a spatial–temporal region in a single frame, which provides a simple yet effective means to enhance grounding performance without incurring significant additional cost. Specifically, we innovatively contribute a spatial–temporal MIL framework. In the temporal-level MIL, by applying Gaussian weighting to the frames of a video, we assign higher weights to the frames that are close to the annotated frame, while lower weights are assigned to frames that are further away. In the spatial-level MIL, we propose regions in the each frame and compute their similarity with the annotated bounding box, selecting regions with higher similarity scores for training. Ultimately, temporal-level and spatial-level MILs are integrated to jointly optimize the accuracy of both types of grounding. Through experimental evaluations on two re-annotated datasets, our proposed framework has been demonstrated to exhibit superiority in terms of both overall performance comparison and detailed micro-level analyses. Compared to the latest weakly-supervised methods on the VidSTG dataset, our method improves the temporal localization performance by at least 10.35% and the spatial localization performance by at least 11.89%.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"314 ","pages":"Article 113200"},"PeriodicalIF":7.2,"publicationDate":"2025-02-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143519214","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Enhanced differential evolution through chaotic and Euclidean models for solving flexible process planning

IF 7.2 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Knowledge-Based Systems

Pub Date : 2025-02-27 DOI: 10.1016/j.knosys.2025.113189

Eduardo H. Haro, Diego Oliva, Luis A. Beltrán, Angel Casas-Ordaz

The Differential Evolution (DE) algorithm is a well-founded technique proposed in 1995. Its simple but effective structure has attracted attention over the years. Moreover, its design has allowed the publication of several variants that are positioned as competitive approaches. Nonetheless, most of these modifications focus on the mutation and crossover stages, while the initialization and selection phases have received less attention. However, in recent years, different studies have demonstrated the advantage of improving these two stages for certain optimization problems. Therefore, in this work, a new DE variant is proposed to solve the Flexible Process Planning (FPP) problem, one of the most relevant tasks in manufacturing optimization. In this work, the DE initialization is enhanced by a chaotic Opposition-Based-Learning (OBL) method. At the same time, the selection is improved by a dynamic model based on Euclidean distances. This work’s novelty lies in optimizing a real-world task with a DE variant focused exclusively on the initialization and selection operators. The efficiency of the method was also tested on the CEC-2017 functions. The outcomes probe its robustness and competitiveness for general optimization and manufacturing systems.

{"title":"Enhanced differential evolution through chaotic and Euclidean models for solving flexible process planning","authors":"Eduardo H. Haro, Diego Oliva, Luis A. Beltrán, Angel Casas-Ordaz","doi":"10.1016/j.knosys.2025.113189","DOIUrl":"10.1016/j.knosys.2025.113189","url":null,"abstract":"<div><div>The Differential Evolution (DE) algorithm is a well-founded technique proposed in 1995. Its simple but effective structure has attracted attention over the years. Moreover, its design has allowed the publication of several variants that are positioned as competitive approaches. Nonetheless, most of these modifications focus on the mutation and crossover stages, while the initialization and selection phases have received less attention. However, in recent years, different studies have demonstrated the advantage of improving these two stages for certain optimization problems. Therefore, in this work, a new DE variant is proposed to solve the Flexible Process Planning (FPP) problem, one of the most relevant tasks in manufacturing optimization. In this work, the DE initialization is enhanced by a chaotic Opposition-Based-Learning (OBL) method. At the same time, the selection is improved by a dynamic model based on Euclidean distances. This work’s novelty lies in optimizing a real-world task with a DE variant focused exclusively on the initialization and selection operators. The efficiency of the method was also tested on the CEC-2017 functions. The outcomes probe its robustness and competitiveness for general optimization and manufacturing systems.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"314 ","pages":"Article 113189"},"PeriodicalIF":7.2,"publicationDate":"2025-02-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143549031","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Semantic relation-aware graph attention network with noise augmented layer-wise contrastive learning for recommendation

IF 7.2 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Knowledge-Based Systems

Pub Date : 2025-02-27 DOI: 10.1016/j.knosys.2025.113217

Jianfang Liu , Wei Wang , Baolin Yi , Huanyu Zhang , Xiaoxuan Shen

Recommender systems based on knowledge graphs enhance the explainability of recommendations by incorporating external knowledge. Nevertheless, the accuracy of recommendations heavily depends on dense interaction data and high-quality knowledge graphs, both of which commonly suffer from data sparsity. Introducing graph contrastive learning to enhance representation quality can effectively improve recommendation performance. Existing graph contrastive learning methods that use graph augmentation can alleviate the data sparsity problem. However, they often neglect the semantic modeling of relation embeddings and lack sufficient contrastive information, leading to insufficient utilization of the embedding space for relations and nodes. To address this, we propose a semantic relation-aware graph attention network with a noise augmented layer-wise contrastive learning model for recommendation, named SRGAN. Specifically, we design a semantic relation-aware graph attention network that updates the semantics of relations during multi-layer iterations to better capture user preferences. Additionally, we construct a noise-augmented layer-wise contrastive learning model, employing simple yet effective noise perturbations to generate contrastive views for entities and relations. By maximizing the consistency of the representations in each layer, the model achieves alignment with the lower-level features of the intermediate layers. Extensive experiments on three public benchmark datasets demonstrate that our proposed method significantly outperforms current approaches. To ensure reproducibility, we make the code and data from our experiments publicly available on https://github.com/liujianfang2021/SRGAN.

{"title":"Semantic relation-aware graph attention network with noise augmented layer-wise contrastive learning for recommendation","authors":"Jianfang Liu , Wei Wang , Baolin Yi , Huanyu Zhang , Xiaoxuan Shen","doi":"10.1016/j.knosys.2025.113217","DOIUrl":"10.1016/j.knosys.2025.113217","url":null,"abstract":"<div><div>Recommender systems based on knowledge graphs enhance the explainability of recommendations by incorporating external knowledge. Nevertheless, the accuracy of recommendations heavily depends on dense interaction data and high-quality knowledge graphs, both of which commonly suffer from data sparsity. Introducing graph contrastive learning to enhance representation quality can effectively improve recommendation performance. Existing graph contrastive learning methods that use graph augmentation can alleviate the data sparsity problem. However, they often neglect the semantic modeling of relation embeddings and lack sufficient contrastive information, leading to insufficient utilization of the embedding space for relations and nodes. To address this, we propose a semantic relation-aware graph attention network with a noise augmented layer-wise contrastive learning model for recommendation, named <em><u>SRGAN</u></em>. Specifically, we design a semantic relation-aware graph attention network that updates the semantics of relations during multi-layer iterations to better capture user preferences. Additionally, we construct a noise-augmented layer-wise contrastive learning model, employing simple yet effective noise perturbations to generate contrastive views for entities and relations. By maximizing the consistency of the representations in each layer, the model achieves alignment with the lower-level features of the intermediate layers. Extensive experiments on three public benchmark datasets demonstrate that our proposed method significantly outperforms current approaches. To ensure reproducibility, we make the code and data from our experiments publicly available on <span><span>https://github.com/liujianfang2021/SRGAN</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"314 ","pages":"Article 113217"},"PeriodicalIF":7.2,"publicationDate":"2025-02-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143534358","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Fully Bayesian differential Gaussian processes through stochastic differential equations

IF 7.2 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Knowledge-Based Systems

Pub Date : 2025-02-25 DOI: 10.1016/j.knosys.2025.113187

Jian Xu , Zhiqi Lin , Min Chen , Junmei Yang , Delu Zeng , John Paisley

Deep Gaussian process models typically employ discrete hierarchies, but recent advancements in differential Gaussian processes (DiffGPs) have extended these models to infinite depths. However, existing DiffGP approaches often overlook the uncertainty in kernel hyperparameters by treating them as fixed and time-invariant, which degrades the model’s predictive performance and neglects the posterior distribution. In this work, we introduce a fully Bayesian framework that models kernel hyperparameters as random variables and utilizes coupled stochastic differential equations (SDEs) to jointly learn their posterior distributions alongside those of inducing points. By incorporating the estimation uncertainty of hyperparameters, our method significantly enhances model flexibility and adaptability to complex dynamic systems. Furthermore, we employ a black-box adaptive SDE solver with a neural network to achieve realistic, time-varying posterior approximations, thereby improving the expressiveness of the variational posterior. Comprehensive experimental evaluations demonstrate that our approach outperforms traditional methods in terms of flexibility, accuracy, and other key performance metrics. This work not only provides a robust Bayesian extension to DiffGP models but also validates its effectiveness in handling intricate dynamic behaviors, thereby advancing the applicability of Gaussian process models in diverse real-world scenarios.

{"title":"Fully Bayesian differential Gaussian processes through stochastic differential equations","authors":"Jian Xu , Zhiqi Lin , Min Chen , Junmei Yang , Delu Zeng , John Paisley","doi":"10.1016/j.knosys.2025.113187","DOIUrl":"10.1016/j.knosys.2025.113187","url":null,"abstract":"<div><div>Deep Gaussian process models typically employ discrete hierarchies, but recent advancements in differential Gaussian processes (DiffGPs) have extended these models to infinite depths. However, existing DiffGP approaches often overlook the uncertainty in kernel hyperparameters by treating them as fixed and time-invariant, which degrades the model’s predictive performance and neglects the posterior distribution. In this work, we introduce a fully Bayesian framework that models kernel hyperparameters as random variables and utilizes coupled stochastic differential equations (SDEs) to jointly learn their posterior distributions alongside those of inducing points. By incorporating the estimation uncertainty of hyperparameters, our method significantly enhances model flexibility and adaptability to complex dynamic systems. Furthermore, we employ a black-box adaptive SDE solver with a neural network to achieve realistic, time-varying posterior approximations, thereby improving the expressiveness of the variational posterior. Comprehensive experimental evaluations demonstrate that our approach outperforms traditional methods in terms of flexibility, accuracy, and other key performance metrics. This work not only provides a robust Bayesian extension to DiffGP models but also validates its effectiveness in handling intricate dynamic behaviors, thereby advancing the applicability of Gaussian process models in diverse real-world scenarios.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"314 ","pages":"Article 113187"},"PeriodicalIF":7.2,"publicationDate":"2025-02-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143509730","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Camera-aware Embedding Refinement for unsupervised person re-identification

IF 7.2 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Knowledge-Based Systems

Pub Date : 2025-02-25 DOI: 10.1016/j.knosys.2025.113195

Yimin Liu , Meibin Qi , Yongle Zhang , Wenbo Xu , Qiang Wu

We propose a data-centric approach, Camera-aware Embedding Refinement (CER), to enhance the discriminability of unsupervised person re-identification. CER consists of two components: camera proxy memory and Camera-aware Embedding Generation (CEG). Camera proxy memory is initialized with original embeddings and updated during training using auxiliary embeddings generated by CEG to ensure consistency within the memory. The auxiliary embeddings are created by CEG based on the intrinsic relationships within the original dataset, accounting for image variations caused by different camera perspectives. Specifically, CEG handles scenarios such as images of the same person captured by different cameras, images of different individuals captured by the same camera, and images of different individuals from different cameras. Training the downstream unsupervised Re-ID model with only the auxiliary embeddings significantly improves feature discriminability. Our method focuses on generating auxiliary embeddings and can be adapted for various unsupervised Re-ID models. Extensive experiments show that our approach consistently outperforms state-of-the-art techniques.

{"title":"Camera-aware Embedding Refinement for unsupervised person re-identification","authors":"Yimin Liu , Meibin Qi , Yongle Zhang , Wenbo Xu , Qiang Wu","doi":"10.1016/j.knosys.2025.113195","DOIUrl":"10.1016/j.knosys.2025.113195","url":null,"abstract":"<div><div>We propose a data-centric approach, Camera-aware Embedding Refinement (CER), to enhance the discriminability of unsupervised person re-identification. CER consists of two components: camera proxy memory and Camera-aware Embedding Generation (CEG). Camera proxy memory is initialized with original embeddings and updated during training using auxiliary embeddings generated by CEG to ensure consistency within the memory. The auxiliary embeddings are created by CEG based on the intrinsic relationships within the original dataset, accounting for image variations caused by different camera perspectives. Specifically, CEG handles scenarios such as images of the same person captured by different cameras, images of different individuals captured by the same camera, and images of different individuals from different cameras. Training the downstream unsupervised Re-ID model with only the auxiliary embeddings significantly improves feature discriminability. Our method focuses on generating auxiliary embeddings and can be adapted for various unsupervised Re-ID models. Extensive experiments show that our approach consistently outperforms state-of-the-art techniques.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"314 ","pages":"Article 113195"},"PeriodicalIF":7.2,"publicationDate":"2025-02-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143534356","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Knowledge assimilation: Implementing knowledge-guided agricultural large language model

IF 7.2 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Knowledge-Based Systems

Pub Date : 2025-02-24 DOI: 10.1016/j.knosys.2025.113197

Jingchi Jiang , Lian Yan , Haifeng Liu , Zhenbo Xia , Haotian Wang , Yang Yang , Yi Guan

Although supervised fine-tuning (SFT) and retrieval-augmented generation (RAG) can help large language models (LLMs) incorporate domain knowledge, they have the following limitations: (1) Data scarcity. There is a severe lack of high-quality data and knowledge bases on dialogue in agriculture. (2) Token-level oversight. Current SFT primarily focuses on fitting general tokens, neglecting agricultural-specific tokens. It leads to omissions of critical information in responses. (3) Sentence-level hurdle. Agricultural queries necessitate sentence-level evidence support from domain knowledge bases, which poses a challenge to precision evidence retrievers. This paper introduces a novel Knowledge-guided Agriculture LLM (KALLM) designed to facilitate multi-task decision-making in agricultural settings. We begin by addressing the data quality issue by establishing an annotation standard and constructing a comprehensive dataset consisting of 220,000 Q&A pairs derived from authoritative agricultural documents. At the token level, we propose a knowledge-coordinated SFT approach that enhances the representation of agriculture-specific tokens by amplifying their significance during the decoding process. At the sentence level, we introduce a self-reflective RAG mechanism based on topic matching to improve the accuracy of evidence retrieval. Experimental results compared with seven competitive open-domain LLMs and the current SFT-RAG pipeline show that our KALLM achieves state-of-the-art performance and is significantly superior to existing generation frameworks in terms of response fluency, accuracy, and domain fidelity.

{"title":"Knowledge assimilation: Implementing knowledge-guided agricultural large language model","authors":"Jingchi Jiang , Lian Yan , Haifeng Liu , Zhenbo Xia , Haotian Wang , Yang Yang , Yi Guan","doi":"10.1016/j.knosys.2025.113197","DOIUrl":"10.1016/j.knosys.2025.113197","url":null,"abstract":"<div><div>Although supervised fine-tuning (SFT) and retrieval-augmented generation (RAG) can help large language models (LLMs) incorporate domain knowledge, they have the following limitations: (1) Data scarcity. There is a severe lack of high-quality data and knowledge bases on dialogue in agriculture. (2) Token-level oversight. Current SFT primarily focuses on fitting general tokens, neglecting agricultural-specific tokens. It leads to omissions of critical information in responses. (3) Sentence-level hurdle. Agricultural queries necessitate sentence-level evidence support from domain knowledge bases, which poses a challenge to precision evidence retrievers. This paper introduces a novel Knowledge-guided Agriculture LLM (KALLM) designed to facilitate multi-task decision-making in agricultural settings. We begin by addressing the data quality issue by establishing an annotation standard and constructing a comprehensive dataset consisting of 220,000 Q&A pairs derived from authoritative agricultural documents. At the token level, we propose a knowledge-coordinated SFT approach that enhances the representation of agriculture-specific tokens by amplifying their significance during the decoding process. At the sentence level, we introduce a self-reflective RAG mechanism based on topic matching to improve the accuracy of evidence retrieval. Experimental results compared with seven competitive open-domain LLMs and the current SFT-RAG pipeline show that our KALLM achieves state-of-the-art performance and is significantly superior to existing generation frameworks in terms of response fluency, accuracy, and domain fidelity.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"314 ","pages":"Article 113197"},"PeriodicalIF":7.2,"publicationDate":"2025-02-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143512221","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

User group-enhanced user feature distribution transfer framework for non-overlapping cross-domain recommendations

IF 7.2 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Knowledge-Based Systems

Pub Date : 2025-02-24 DOI: 10.1016/j.knosys.2025.113186

Xiaoying Gao, Ling Ding, Jianting Chen, Yunxiao Yang, Yang Xiang

Cross-domain recommendation (CDR) aims to alleviate data sparsity in the target domain by leveraging rich information from source domains. Most existing approaches rely on overlapping information to transfer knowledge, but in real scenarios, these correspondences are often unknown. This makes it critical to develop CDR systems without overlapping information. However, such CDR systems still face user feature bias between domains and ignore the importance of sparse interaction information from the target domain, resulting in sub-optimal recommendations performance. To address challenges, we propose a User Group-enhanced User Feature Distribution Transfer framework (UGUFDT) for CDR. Specifically, it first utilizes a User Feature Separation Network bridges domains by constructing a cross-domain user–cluster graph to capture transferable user features, while User Feature Reconstructor refines unbiased user representations through reconstruction factors to build an inverse user–cluster graph, filter out source domain-specific noises. Then, we introduce three types of loss function – Difference Loss, Similarity Loss, Reconstruction Loss – to reduce feature distribution discrepancies between domains. Furthermore, to fully exploit interactions in target domain, we propose a User–Group Graph with a Soft Allocation Mechanism, which aggregates group-level preferences to enhance user representations. Finally, a Prediction Layer with a Fusion Mechanism integrates both cross-domain transferable knowledge and target-domain preferences to generate more accurate recommendations. Experiments on three publicly available datasets – ML, AB, and AM – demonstrate that the proposed model significantly outperforms state-of-the-art models on the HR and NDCG evaluation metrics, validating the effectiveness of our model.

{"title":"User group-enhanced user feature distribution transfer framework for non-overlapping cross-domain recommendations","authors":"Xiaoying Gao, Ling Ding, Jianting Chen, Yunxiao Yang, Yang Xiang","doi":"10.1016/j.knosys.2025.113186","DOIUrl":"10.1016/j.knosys.2025.113186","url":null,"abstract":"<div><div>Cross-domain recommendation (CDR) aims to alleviate data sparsity in the target domain by leveraging rich information from source domains. Most existing approaches rely on overlapping information to transfer knowledge, but in real scenarios, these correspondences are often unknown. This makes it critical to develop CDR systems without overlapping information. However, such CDR systems still face user feature bias between domains and ignore the importance of sparse interaction information from the target domain, resulting in sub-optimal recommendations performance. To address challenges, we propose a <strong>U</strong>ser <strong>G</strong>roup-enhanced <strong>U</strong>ser <strong>F</strong>eature <strong>D</strong>istribution <strong>T</strong>ransfer framework (UGUFDT) for CDR. Specifically, it first utilizes a User Feature Separation Network bridges domains by constructing a cross-domain user–cluster graph to capture transferable user features, while User Feature Reconstructor refines unbiased user representations through reconstruction factors to build an inverse user–cluster graph, filter out source domain-specific noises. Then, we introduce three types of loss function – <em>Difference Loss</em>, <em>Similarity Loss</em>, <em>Reconstruction Loss</em> – to reduce feature distribution discrepancies between domains. Furthermore, to fully exploit interactions in target domain, we propose a User–Group Graph with a Soft Allocation Mechanism, which aggregates group-level preferences to enhance user representations. Finally, a Prediction Layer with a Fusion Mechanism integrates both cross-domain transferable knowledge and target-domain preferences to generate more accurate recommendations. Experiments on three publicly available datasets – ML, AB, and AM – demonstrate that the proposed model significantly outperforms state-of-the-art models on the HR and NDCG evaluation metrics, validating the effectiveness of our model.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"314 ","pages":"Article 113186"},"PeriodicalIF":7.2,"publicationDate":"2025-02-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143487819","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Spiking neural network classification of X-ray chest images

IF 7.2 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Knowledge-Based Systems

Pub Date : 2025-02-24 DOI: 10.1016/j.knosys.2025.113194

Marco Gatti, Jessica Amianto Barbato, Claudio Zandron

Spiking Neural Networks (SNNs) are powerful and biologically plausible models of neural processing and represent a transition to a new generation of neural networks, as they address the problem of high resource requirements by significantly reducing energy consumption. In this paper we investigate the use of SNNs for the diagnosis of COVID-19 cases from chest x-rays, by proposing a simple Spiking Neural Network (SNN) that proves to be effective despite the low resources requested with respect to other solutions proposed in the literature. The paper explains the architecture of the SNN and evaluates the performance of the model in terms of both result accuracy and energy consumption. Experimental results show competitive performance in terms of accuracy and a significant reduction in energy consumption.

引用次数: 0