Pub Date : 2026-01-27DOI: 10.1016/j.ins.2026.123132
Yanan Song, Xiangyuan Chen, Ronghua Xu
The rapid advancement of deep learning-based generative technologies has led to remarkable achievements in deepfake applications using video and image media, particularly in areas such as face swapping and expression transfer. However, these developments have also triggered significant concerns regarding media authenticity and information security. Deepfake content often exhibits various artifacts: in the spatial domain, it may suffer from over-smoothed textures, loss of edge details, or jagged distortions; in the frequency domain, abnormal peaks in high-frequency spectra or noise-induced distortions may appear; and at the semantic level, misaligned keypoints and poor temporal coherence are frequently observed. To address these limitations, this study proposes a network architecture that first performs hybrid-domain feature extraction on deepfake samples. The Xception backbone, optimized through a knowledge distillation strategy to remove redundant layers, is combined with the lightweight MesoNet4 architecture to form a dual-branch backbone that can capture semantic features at different levels. While preserving semantic representation, the overall model size is compressed to just 8.0M parameters, achieving both high-precision detection of deepfake samples (accuracy 99%) and real-time inference performance (single-frame latency 10 ms).
{"title":"Dual-branch Meso-Xception network for hybrid-domain feature of deepfake detection","authors":"Yanan Song, Xiangyuan Chen, Ronghua Xu","doi":"10.1016/j.ins.2026.123132","DOIUrl":"10.1016/j.ins.2026.123132","url":null,"abstract":"<div><div>The rapid advancement of deep learning-based generative technologies has led to remarkable achievements in deepfake applications using video and image media, particularly in areas such as face swapping and expression transfer. However, these developments have also triggered significant concerns regarding media authenticity and information security. Deepfake content often exhibits various artifacts: in the spatial domain, it may suffer from over-smoothed textures, loss of edge details, or jagged distortions; in the frequency domain, abnormal peaks in high-frequency spectra or noise-induced distortions may appear; and at the semantic level, misaligned keypoints and poor temporal coherence are frequently observed. To address these limitations, this study proposes a network architecture that first performs hybrid-domain feature extraction on deepfake samples. The Xception backbone, optimized through a knowledge distillation strategy to remove redundant layers, is combined with the lightweight MesoNet4 architecture to form a dual-branch backbone that can capture semantic features at different levels. While preserving semantic representation, the overall model size is compressed to just 8.0M parameters, achieving both high-precision detection of deepfake samples (accuracy <span><math><mo>≥</mo></math></span> 99%) and real-time inference performance (single-frame latency <span><math><mo>≤</mo></math></span> 10 ms).</div></div>","PeriodicalId":51063,"journal":{"name":"Information Sciences","volume":"739 ","pages":"Article 123132"},"PeriodicalIF":6.8,"publicationDate":"2026-01-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146081919","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}
Magnetic resonance (MR) imaging plays a critical role in diagnostic workflows, yet its reliability is frequently compromised by scanner-dependent bias, contrast variability, and intensity drift. Although deep learning methods achieve high performance, they generally require extensive supervision and demonstrate limited robustness across diverse clinical settings.
To address these challenges, we propose a transparent, geometry-aware framework for annotation-free MR enhancement based on -Bézier curves. This model incorporates an adaptive deformation parameter that modulates local curvature, facilitating flexible adaptation to complex anatomical boundaries. The framework comprises three principal mechanisms: (i) adaptive for local responsiveness, (ii) monotone -Bézier tone curves for intensity standardization, and (iii) Tikhonov-regularized optimization for smooth mapping. As a result, the operator remains interpretable, operates in linear time, and provides explicit control over smoothness.
The proposed approach was validated across five public cohorts (BraTS, ACDC, PROMISE12, fastMRI, IXI), demonstrating significant improvements in image fidelity (SSIM, CNR, NIQE) and downstream segmentation accuracy (Dice, HD95) relative to variational filters and state-of-the-art foundation models. Additionally, cross-vendor experiments confirm its robustness without the need for retraining. Collectively, these findings establish -Bézier modeling as a principled, lightweight, and clinically interpretable alternative that complements deep learning by providing a geometry-aware pathway to robust MR representation.
{"title":"Interpretable structural modeling of MR images using q−Bézier curves: A geometry-aware paradigm beyond deep learning","authors":"Faruk Özger , Aytuğ Onan , Nezihe Turhan , Zeynep Ödemiş Özger","doi":"10.1016/j.ins.2026.123147","DOIUrl":"10.1016/j.ins.2026.123147","url":null,"abstract":"<div><div>Magnetic resonance (MR) imaging plays a critical role in diagnostic workflows, yet its reliability is frequently compromised by scanner-dependent bias, contrast variability, and intensity drift. Although deep learning methods achieve high performance, they generally require extensive supervision and demonstrate limited robustness across diverse clinical settings.</div><div>To address these challenges, we propose a transparent, <strong>geometry-aware framework</strong> for annotation-free MR enhancement based on <span><math><mi>q</mi></math></span><strong>-Bézier curves</strong>. This model incorporates an <strong>adaptive deformation parameter</strong> <span><math><mi>q</mi><mo>(</mo><mi>x</mi><mo>)</mo></math></span> that modulates local curvature, facilitating flexible adaptation to complex anatomical boundaries. The framework comprises three principal mechanisms: (i) adaptive <span><math><mi>q</mi><mo>(</mo><mi>x</mi><mo>)</mo></math></span> for local responsiveness, (ii) monotone <span><math><mi>q</mi></math></span>-Bézier tone curves for intensity standardization, and (iii) <strong>Tikhonov-regularized</strong> optimization for smooth mapping. As a result, the operator remains interpretable, operates in linear time, and provides explicit control over smoothness.</div><div>The proposed approach was validated across five public cohorts (BraTS, ACDC, PROMISE12, fastMRI, IXI), demonstrating significant improvements in image fidelity (SSIM, CNR, NIQE) and downstream segmentation accuracy (Dice, HD95) relative to variational filters and state-of-the-art foundation models. Additionally, cross-vendor experiments confirm its <strong>robustness without the need for retraining</strong>. Collectively, these findings establish <span><math><mi>q</mi></math></span>-Bézier modeling as a principled, lightweight, and clinically interpretable alternative that complements deep learning by providing a geometry-aware pathway to robust MR representation.</div></div>","PeriodicalId":51063,"journal":{"name":"Information Sciences","volume":"739 ","pages":"Article 123147"},"PeriodicalIF":6.8,"publicationDate":"2026-01-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146057592","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}
Pub Date : 2026-01-27DOI: 10.1016/j.ins.2026.123135
Dongsheng Ruan , Yuan Zheng , Lide Mu , Ao Ran , Lei Pan , Mingfeng Jiang , Chengjin Yu , Nenggan Zheng , Huafeng Liu
Window-based Transformers have achieved remarkable results in image super-resolution (SR). State-of-the-art SR models generally employ a window self-attention mechanism combined with a multi-layer perceptron (MLP) to effectively capture long-range dependencies. However, the window design and the MLP’s deficiency in capturing spatial dependencies restrict their capacity to utilize global contextual information in images. This paper aims to address this limitation by introducing global context modeling. Specifically, we propose a general global context-injected framework for window self-attention. Within this framework, we develop a new instantiation with a novel global context-injected (GCI) module, which allows each window to take advantage of the contextual information from other windows. The GCI module is lightweight and can be easily integrated into existing window-based Transformers, improving performance with negligible increases in parameters and computational costs. Furthermore, we introduce a window self-attention (WSA) to vision state space (VSS) flow to further enhance the ability for global context modeling. We incorporate our advancements into popular SR models, such as SwinIR and SRFormer, creating enhanced versions. Extensive experiments on three representative SR tasks demonstrate the effectiveness of our methods, showing substantial performance improvements over their vanilla counterparts. Notably, our GCI-MSRformer outperforms current state-of-the-art models like MambaIR.
{"title":"Global context modeling for image super-resolution transformer","authors":"Dongsheng Ruan , Yuan Zheng , Lide Mu , Ao Ran , Lei Pan , Mingfeng Jiang , Chengjin Yu , Nenggan Zheng , Huafeng Liu","doi":"10.1016/j.ins.2026.123135","DOIUrl":"10.1016/j.ins.2026.123135","url":null,"abstract":"<div><div>Window-based Transformers have achieved remarkable results in image super-resolution (SR). State-of-the-art SR models generally employ a window self-attention mechanism combined with a multi-layer perceptron (MLP) to effectively capture long-range dependencies. However, the window design and the MLP’s deficiency in capturing spatial dependencies restrict their capacity to utilize global contextual information in images. This paper aims to address this limitation by introducing global context modeling. Specifically, we propose a general global context-injected framework for window self-attention. Within this framework, we develop a new instantiation with a novel global context-injected (GCI) module, which allows each window to take advantage of the contextual information from other windows. The GCI module is lightweight and can be easily integrated into existing window-based Transformers, improving performance with negligible increases in parameters and computational costs. Furthermore, we introduce a window self-attention (WSA) to vision state space (VSS) flow to further enhance the ability for global context modeling. We incorporate our advancements into popular SR models, such as SwinIR and SRFormer, creating enhanced versions. Extensive experiments on three representative SR tasks demonstrate the effectiveness of our methods, showing substantial performance improvements over their vanilla counterparts. Notably, our GCI-MSRformer outperforms current state-of-the-art models like MambaIR.</div></div>","PeriodicalId":51063,"journal":{"name":"Information Sciences","volume":"739 ","pages":"Article 123135"},"PeriodicalIF":6.8,"publicationDate":"2026-01-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146081918","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}
Pub Date : 2026-01-27DOI: 10.1016/j.ins.2026.123162
Peide Liu , Yurong Qian , Ran Dang , Fei Teng , Peng Wang
As decision-making problems continue to expand, group decision-making (GDM) has seen growing interest in the application of social trust networks (STN). The trust capacity stems of a decision maker (DM) from both self-trust and external social support. Misalignment between the two may lead to trust crises or overconfidence, affecting decision outcomes. Under a two-dimensional linguistic (2DL) environment, a consensus method for multi-attribute group decision-making (MAGDM) that combines internal and external trust mechanisms is presented in this paper. First, DMs’ self-trust is assessed through subjective judgment in the 2DL setting, and then compared with social trust support from the STN in a two-stage trust risk evaluation to align individual competence with external expectations. Next, individual opinions are optimized during the consensus process while managing trust risk. In attribute assignment, individual weights are determined through the interaction between DMs and STN, and attribute weights are calculated using information entropy. To better capture DMs’ psychological behavior, such as regret and hesitation during comparisons, a new MAGDM ranking method integrating regret theory and the SIR method is proposed to improve decision reliability. Lastly, through an illustrative application in an actual data element market context and comparative analysis, the effectiveness of the proposed method is demonstrated, offering actionable insights to support decisions pertaining to data factor market development.
{"title":"Multi-attribute group consensus decision-making with two-stage trust risk adjustment","authors":"Peide Liu , Yurong Qian , Ran Dang , Fei Teng , Peng Wang","doi":"10.1016/j.ins.2026.123162","DOIUrl":"10.1016/j.ins.2026.123162","url":null,"abstract":"<div><div>As decision-making problems continue to expand, group decision-making (GDM) has seen growing interest in the application of social trust networks (STN). The trust capacity stems of a decision maker (DM) from both self-trust and external social support. Misalignment between the two may lead to trust crises or overconfidence, affecting decision outcomes. Under a two-dimensional linguistic (2DL) environment, a consensus method for multi-attribute group decision-making (MAGDM) that combines internal and external trust mechanisms is presented in this paper. First, DMs’ self-trust is assessed through subjective judgment in the 2DL setting, and then compared with social trust support from the STN in a two-stage trust risk evaluation to align individual competence with external expectations. Next, individual opinions are optimized during the consensus process while managing trust risk. In attribute assignment, individual weights are determined through the interaction between DMs and STN, and attribute weights are calculated using information entropy. To better capture DMs’ psychological behavior, such as regret and hesitation during comparisons, a new MAGDM ranking method integrating regret theory and the SIR method is proposed to improve decision reliability. Lastly, through an illustrative application in an actual data element market context and comparative analysis, the effectiveness of the proposed method is demonstrated, offering actionable insights to support decisions pertaining to data factor market development.</div></div>","PeriodicalId":51063,"journal":{"name":"Information Sciences","volume":"739 ","pages":"Article 123162"},"PeriodicalIF":6.8,"publicationDate":"2026-01-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146191053","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}
Pub Date : 2026-01-27DOI: 10.1016/j.ins.2026.123148
Witold Andrzejewski, Pawel Boinski
In this paper, we investigate mining Mixed-Drove spatio-temporal Co-Occurrence Patterns (MDCOPs). MDCOPs represent sets of object types frequently located together for a given minimum fraction of time. Current solutions fail to address several important factors in practical applications. Specifically, state-of-the-art methods rely on a series of snapshots, i.e., discrete object positions recorded at predefined timestamps rather than their trajectories. However, spatio-temporal data gathering often depends on unsynchronized distributed sensors that independently register positions for each object.
To tackle this issue using traditional methods, one can interpolate object positions at snapshot timestamps. However, this raises another challenge: determining the optimal number of snapshots while balancing accuracy, processing time, and memory requirements. To overcome these limitations, we formulate a generalized MDCOP mining problem and introduce GMDCOP-Miner, an algorithm that employs a new, generalized time-prevalence measure. The proposed algorithm provides the most accurate results, equal to those obtained via state-of-the-art methods with the number of snapshots tending to infinity. Moreover, our experiments demonstrate that GMDCOP-Miner surpasses existing approaches in both processing time and memory efficiency.
{"title":"Generalized mining of mixed drove co-occurrence patterns","authors":"Witold Andrzejewski, Pawel Boinski","doi":"10.1016/j.ins.2026.123148","DOIUrl":"10.1016/j.ins.2026.123148","url":null,"abstract":"<div><div>In this paper, we investigate mining Mixed-Drove spatio-temporal Co-Occurrence Patterns (MDCOPs). MDCOPs represent sets of object types frequently located together for a given minimum fraction of time. Current solutions fail to address several important factors in practical applications. Specifically, state-of-the-art methods rely on a series of snapshots, i.e., discrete object positions recorded at predefined timestamps rather than their trajectories. However, spatio-temporal data gathering often depends on unsynchronized distributed sensors that independently register positions for each object.</div><div>To tackle this issue using traditional methods, one can interpolate object positions at snapshot timestamps. However, this raises another challenge: determining the optimal number of snapshots while balancing accuracy, processing time, and memory requirements. To overcome these limitations, we formulate a generalized MDCOP mining problem and introduce GMDCOP-Miner, an algorithm that employs a new, generalized time-prevalence measure. The proposed algorithm provides the most accurate results, equal to those obtained via state-of-the-art methods with the number of snapshots tending to infinity. Moreover, our experiments demonstrate that GMDCOP-Miner surpasses existing approaches in both processing time and memory efficiency.</div></div>","PeriodicalId":51063,"journal":{"name":"Information Sciences","volume":"739 ","pages":"Article 123148"},"PeriodicalIF":6.8,"publicationDate":"2026-01-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146190947","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}
Pub Date : 2026-01-24DOI: 10.1016/j.ins.2026.123139
Jingbin Wang, Yumeng Zhang, Zeyuan Lin, Jinsong Lai, Kun Guo
Temporal knowledge graphs (TKGs) comprise timestamped facts and are widely used in intelligent systems. However, large-scale TKGs are often incomplete, therefore Temporal Knowledge Graph Completion (TKGC) is an important task. Existing approaches mostly use local neighborhoods to learn entity and relation representations, ignoring query-aware global semantics and semantic linkages between quadruples. Furthermore, timestamps are frequently considered as independent features, ignoring their periodicity and interactions with the graph structure. We propose T-GRIN (Temporal Graph completion via Representation and INteraction) to incorporate query-aware global semantic representations and deep interaction between entities and relations. T-GRIN employs a dynamic time encoder to capture periodic temporal patterns, an entity encoder with relation-enhanced mechanisms to highlight query-relevant contexts, and a relation encoder with multi-head attention to model diverse semantics under temporal and entity contexts. Furthermore, an interactive convolutional decoder is designed to improve feature interaction and high-order semantic composition. Extensive experiments on benchmark datasets demonstrate the effectiveness of T-GRIN. In ICEWS05-15, T-GRIN outperforms the previous best model by 8.9% MRR and 10.9% Hit@1.
{"title":"Temporal knowledge graph completion via global structural representation and deep interaction","authors":"Jingbin Wang, Yumeng Zhang, Zeyuan Lin, Jinsong Lai, Kun Guo","doi":"10.1016/j.ins.2026.123139","DOIUrl":"10.1016/j.ins.2026.123139","url":null,"abstract":"<div><div>Temporal knowledge graphs (TKGs) comprise timestamped facts and are widely used in intelligent systems. However, large-scale TKGs are often incomplete, therefore Temporal Knowledge Graph Completion (TKGC) is an important task. Existing approaches mostly use local neighborhoods to learn entity and relation representations, ignoring query-aware global semantics and semantic linkages between quadruples. Furthermore, timestamps are frequently considered as independent features, ignoring their periodicity and interactions with the graph structure. We propose <strong>T-GRIN</strong> (<strong>T</strong>emporal <strong>G</strong>raph completion via <strong>R</strong>epresentation and <strong>IN</strong>teraction) to incorporate query-aware global semantic representations and deep interaction between entities and relations. T-GRIN employs a dynamic time encoder to capture periodic temporal patterns, an entity encoder with relation-enhanced mechanisms to highlight query-relevant contexts, and a relation encoder with multi-head attention to model diverse semantics under temporal and entity contexts. Furthermore, an interactive convolutional decoder is designed to improve feature interaction and high-order semantic composition. Extensive experiments on benchmark datasets demonstrate the effectiveness of T-GRIN. In ICEWS05-15, T-GRIN outperforms the previous best model by 8.9% MRR and 10.9% Hit@1.</div></div>","PeriodicalId":51063,"journal":{"name":"Information Sciences","volume":"739 ","pages":"Article 123139"},"PeriodicalIF":6.8,"publicationDate":"2026-01-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146081915","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}
Pub Date : 2026-01-24DOI: 10.1016/j.ins.2026.123136
Yanfei Sun , Junyu Wang , Rui Yin
Nighttime rainy conditions severely degrade visual quality in applications such as autonomous driving and aerial surveillance, where images suffer from compounded low-light and rain degradations. Diffusion models offer strong generative priors but face limitations in image restoration, including poor controllability, structural distortion, and domain gaps with degraded images. We present MR-SDformer, a novel framework that integrates Retinex-based decomposition with diffusion priors for joint nighttime deraining and low-light enhancement. The key innovation is the Maximum-Value Retinex decomposition, which isolates high-intensity rain streaks into the illumination map and produces a rain-free reflectance map that faithfully preserves intrinsic scene content. This decomposition not only bridges the gap between rainy inputs and rain-free priors but also provides complementary guidance to the generative process. Building on this, we design an asymmetric Hybrid Conditional Transformer that leverages the decomposed illumination and reflectance maps to condition the frozen diffusion model more effectively, enabling precise multi-modal feature fusion and high-fidelity reconstruction. Extensive experiments on both synthetic and real-world datasets confirm that MR-SDformer achieves state-of-the-art performance, delivering clearer structure, enhanced illumination, and more realistic visual quality under nighttime rainy conditions.
{"title":"Maximum-Value retinex decomposition guided generative priors for joint deraining and low-light image enhancement","authors":"Yanfei Sun , Junyu Wang , Rui Yin","doi":"10.1016/j.ins.2026.123136","DOIUrl":"10.1016/j.ins.2026.123136","url":null,"abstract":"<div><div>Nighttime rainy conditions severely degrade visual quality in applications such as autonomous driving and aerial surveillance, where images suffer from compounded low-light and rain degradations. Diffusion models offer strong generative priors but face limitations in image restoration, including poor controllability, structural distortion, and domain gaps with degraded images. We present MR-SDformer, a novel framework that integrates Retinex-based decomposition with diffusion priors for joint nighttime deraining and low-light enhancement. The key innovation is the Maximum-Value Retinex decomposition, which isolates high-intensity rain streaks into the illumination map and produces a rain-free reflectance map that faithfully preserves intrinsic scene content. This decomposition not only bridges the gap between rainy inputs and rain-free priors but also provides complementary guidance to the generative process. Building on this, we design an asymmetric Hybrid Conditional Transformer that leverages the decomposed illumination and reflectance maps to condition the frozen diffusion model more effectively, enabling precise multi-modal feature fusion and high-fidelity reconstruction. Extensive experiments on both synthetic and real-world datasets confirm that MR-SDformer achieves state-of-the-art performance, delivering clearer structure, enhanced illumination, and more realistic visual quality under nighttime rainy conditions.</div></div>","PeriodicalId":51063,"journal":{"name":"Information Sciences","volume":"739 ","pages":"Article 123136"},"PeriodicalIF":6.8,"publicationDate":"2026-01-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146081928","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}
Efficient and scalable path planning is a critical challenge for autonomous robotic systems, particularly in complex real-world scenarios. Traditional heuristic search algorithms like A often struggle with scalability and adaptability in such environments. To address these limitations, we improve a search framework that integrates learned, instance-specific heuristics with conventional pathfinding techniques. Leveraging autoencoder transformer networks, we predict two key heuristic functions—Correction Factor (CF) and Path Probability Map (PPM)—trained on diverse datasets—the Motion Planning (MP) and Tiled-MP datasets—to cover a wide range of path planning scenarios. When integrated with Weighted A (WA) algorithm, this approach optimally solves 88% of MP instances, with paths averaging less than 0.7% longer than optimal, and requiring nearly five times fewer node expansions. The framework demonstrates the advantages of heuristic learning in handling larger path planning problems, with inference time accounting for just 10% of the total search duration. It solves nearly half of the most complex instances optimally, showcasing strong scalability for real-time robotics applications. The framework performs well in unseen environments, solving over 25% of new problems perfectly, finding near-optimal solutions with paths less than 7% longer than optimal, and requiring fewer than two-thirds of the typical expansions. Our framework outperforms learnable planners in both scalability and generalization.
{"title":"Scalable and generalizable path planning for robotic navigation using transformer-based heuristic learning","authors":"Elie Thellier, Adolfo Perrusquía, Antonios Tsourdos","doi":"10.1016/j.ins.2026.123149","DOIUrl":"10.1016/j.ins.2026.123149","url":null,"abstract":"<div><div>Efficient and scalable path planning is a critical challenge for autonomous robotic systems, particularly in complex real-world scenarios. Traditional heuristic search algorithms like A<span><math><msup><mspace></mspace><mo>∗</mo></msup></math></span> often struggle with scalability and adaptability in such environments. To address these limitations, we improve a search framework that integrates learned, instance-specific heuristics with conventional pathfinding techniques. Leveraging autoencoder transformer networks, we predict two key heuristic functions—Correction Factor (CF) and Path Probability Map (PPM)—trained on diverse datasets—the Motion Planning (MP) and Tiled-MP datasets—to cover a wide range of path planning scenarios. When integrated with Weighted A<span><math><msup><mspace></mspace><mo>∗</mo></msup></math></span> (WA<span><math><msup><mspace></mspace><mo>∗</mo></msup></math></span>) algorithm, this approach optimally solves 88% of MP instances, with paths averaging less than 0.7% longer than optimal, and requiring nearly five times fewer node expansions. The framework demonstrates the advantages of heuristic learning in handling larger path planning problems, with inference time accounting for just 10% of the total search duration. It solves nearly half of the most complex instances optimally, showcasing strong scalability for real-time robotics applications. The framework performs well in unseen environments, solving over 25% of new problems perfectly, finding near-optimal solutions with paths less than 7% longer than optimal, and requiring fewer than two-thirds of the typical expansions. Our framework outperforms learnable planners in both scalability and generalization.</div></div>","PeriodicalId":51063,"journal":{"name":"Information Sciences","volume":"739 ","pages":"Article 123149"},"PeriodicalIF":6.8,"publicationDate":"2026-01-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146081916","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}
Pub Date : 2026-01-24DOI: 10.1016/j.ins.2026.123146
Junyi Gou , Liangliang Sun , Jing Liu , Zhenghao Song , Ge Guo , She-Gan Gao , Ke Liu , Natalja Matsveichuk , Yuri Sotskov
Differential Evolution (DE), a population-driven stochastic optimization technique, has garnered significant interest among researchers across diverse disciplines because of its simple approach, high resilience, and few control parameters. However, numerous existing DE variants frequently encounter limitations when tackling intricate optimization problems, especially due to premature convergence weakness. To mitigate these deficiencies, the paper proposes an adaptive differential evolution with a deeply informed mutation strategy and historical information for numerical optimization (ADEDH), the main contributions of which can be outlined as follows: Firstly, a bi-stage parameter control strategy is proposed to achieve a better balance between exploration and exploitation. Secondly, a deeply informed mutation strategy is implemented, which uses the historical population to mirror the objective landscape and help guide the evolution. Thirdly, a diversity enhancement strategy based on historical information is proposed to tackle the premature convergence weakness. ADEDH is evaluated against nine outstanding competitors under a vast testing framework, containing CEC2013, CEC2014, and CEC2017 test suites. Additionally, the feasibility of ADEDH is further validated through its application to the parameter identification problem of a photovoltaic model. Experimental results demonstrate that ADEDH diversifies the population, attains superior solution precision, and achieves better stability.
{"title":"An adaptive differential evolution with deeply informed mutation strategy and historical information for numerical optimization","authors":"Junyi Gou , Liangliang Sun , Jing Liu , Zhenghao Song , Ge Guo , She-Gan Gao , Ke Liu , Natalja Matsveichuk , Yuri Sotskov","doi":"10.1016/j.ins.2026.123146","DOIUrl":"10.1016/j.ins.2026.123146","url":null,"abstract":"<div><div>Differential Evolution (DE), a population-driven stochastic optimization technique, has garnered significant interest among researchers across diverse disciplines because of its simple approach, high resilience, and few control parameters. However, numerous existing DE variants frequently encounter limitations when tackling intricate optimization problems, especially due to premature convergence weakness. To mitigate these deficiencies, the paper proposes an adaptive differential evolution with a deeply informed mutation strategy and historical information for numerical optimization (ADEDH), the main contributions of which can be outlined as follows: Firstly, a bi-stage parameter control strategy is proposed to achieve a better balance between exploration and exploitation. Secondly, a deeply informed mutation strategy is implemented, which uses the historical population to mirror the objective landscape and help guide the evolution. Thirdly, a diversity enhancement strategy based on historical information is proposed to tackle the premature convergence weakness. ADEDH is evaluated against nine outstanding competitors under a vast testing framework, containing CEC2013, CEC2014, and CEC2017 test suites. Additionally, the feasibility of ADEDH is further validated through its application to the parameter identification problem of a photovoltaic model. Experimental results demonstrate that ADEDH diversifies the population, attains superior solution precision, and achieves better stability.</div></div>","PeriodicalId":51063,"journal":{"name":"Information Sciences","volume":"739 ","pages":"Article 123146"},"PeriodicalIF":6.8,"publicationDate":"2026-01-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146191057","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}
In today’s economic climate, enterprises face a variety of internal and external risks, making bankruptcy prediction critical for risk management. The traditional statistical and machine learning-based methods mainly rely on economic indicators, which are insufficient for deducing the risk propagation among enterprises. Recently, researchers have begun to explore the use of graph neural networks, utilizing their message-passing mechanisms to simulate the risk propagation process. However, existing graph-based methods often neglect degree imbalances, leading to high misjudgment rates for sparsely connected nodes. Furthermore, existing methods typically use a risk-oriented decision model to evaluate the likelihood of bankruptcy, which may lead to the overestimation of bankruptcy probabilities.
To address these issues, we propose a novel bankruptcy prediction model which consists of several key components, including a data-driven explicit risk encoding module, a global multihead attention-based implicit risk encoding module, a hierarchical hypergraph-based external risk enhancement module, and a dual-decision expert-based risk assessment module. We extend the traditional graph structure to a hierarchical hypergraph structure and design a corresponding information propagation strategy to alleviate the degree imbalance issue. Furthermore, a dual-decision assessment module is designed to integrate the perspectives of both risk and non-risk experts to prevent the overestimation of bankruptcy probabilities. Extensive experiments conducted on a real-world dataset demonstrate the effectiveness of the proposed model, which achieves an accuracy of 77.64% and an AUC of 0.8270, significantly outperforming existing methods.
{"title":"BPHD: Enterprise bankruptcy prediction with a hierarchical hypergraph and dual-decision experts","authors":"Boyuan Ren , Hongrui Guo , Hongzhi Liu , Xudong Tang , Jingming Xue , Zhonghai Wu","doi":"10.1016/j.ins.2026.123142","DOIUrl":"10.1016/j.ins.2026.123142","url":null,"abstract":"<div><div>In today’s economic climate, enterprises face a variety of internal and external risks, making bankruptcy prediction critical for risk management. The traditional statistical and machine learning-based methods mainly rely on economic indicators, which are insufficient for deducing the risk propagation among enterprises. Recently, researchers have begun to explore the use of graph neural networks, utilizing their message-passing mechanisms to simulate the risk propagation process. However, existing graph-based methods often neglect degree imbalances, leading to high misjudgment rates for sparsely connected nodes. Furthermore, existing methods typically use a risk-oriented decision model to evaluate the likelihood of bankruptcy, which may lead to the overestimation of bankruptcy probabilities.</div><div>To address these issues, we propose a novel bankruptcy prediction model which consists of several key components, including a data-driven explicit risk encoding module, a global multihead attention-based implicit risk encoding module, a hierarchical hypergraph-based external risk enhancement module, and a dual-decision expert-based risk assessment module. We extend the traditional graph structure to a hierarchical hypergraph structure and design a corresponding information propagation strategy to alleviate the degree imbalance issue. Furthermore, a dual-decision assessment module is designed to integrate the perspectives of both risk and non-risk experts to prevent the overestimation of bankruptcy probabilities. Extensive experiments conducted on a real-world dataset demonstrate the effectiveness of the proposed model, which achieves an accuracy of 77.64% and an AUC of 0.8270, significantly outperforming existing methods.</div></div>","PeriodicalId":51063,"journal":{"name":"Information Sciences","volume":"738 ","pages":"Article 123142"},"PeriodicalIF":6.8,"publicationDate":"2026-01-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146081294","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}
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号