Applied Soft Computing最新文献

{"title":"Bi-directional distillation enabling progressive lightweight model training in edge computing-based federated learning","authors":"Yishan Chen ,&nbsp;Wenshuo Dai ,&nbsp;Zhen Qin ,&nbsp;Junxiao Han","doi":"10.1016/j.asoc.2026.114712","DOIUrl":"10.1016/j.asoc.2026.114712","url":null,"abstract":"<div><div>Wireless Federated Learning (FL) has emerged as a promising paradigm for enabling collaborative intelligence in edge computing while preserving data privacy by exchanging model parameters instead of raw data. However, deploying large-scale models remains challenging due to the limited computational resources and constrained communication bandwidth of edge devices. To address this issue, the paper proposes a Bi-Directional Distillation-Driven Hierarchical Framework (BDH-LM) that enables Progressive Lightweight Model Training across edge and cloud. Central to our framework is the Streamline Teacher Network (STN) algorithm, which progressively compresses the network size by generating a series of compact student models that replace large-scale models without significant performance degradation. A bi-directional knowledge distillation mechanism is established to facilitate mutual knowledge exchange between the cloud and edge models, thereby reducing the computational burden on cloud servers while preserving user privacy on edge devices. To further minimize communication overhead, a data-value-aware selection strategy is introduced to identify and transfer only the most informative samples. Moreover, we formulate a joint optimization problem that considers energy and latency constraints, optimizing pruning ratios, CPU frequencies, uplink power, and bandwidth allocation to accelerate the training process under resource limitations. Finally, experiments conducted on CIFAR-10, CIFAR-100, and SVHN datasets further validate the theoretical results.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"191 ","pages":"Article 114712"},"PeriodicalIF":6.6,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146080102","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}

{"title":"HQML-NLP: A hybrid quantum machine learning framework for scholarly AI-text detection","authors":"Layth Rafea Hazim ,&nbsp;Oguz Ata","doi":"10.1016/j.asoc.2026.114634","DOIUrl":"10.1016/j.asoc.2026.114634","url":null,"abstract":"<div><div>The swift growth of the gargantuan language models has made it even harder to tell apart writings done by humans and AI especially in academia which led to the need for the setup of trustworthy detection frameworks that will satisfactorily balance accuracy, interpretability, and efficiency. In this paper, we walk you through the HQML-NLP, a detection system using a hybrid quantum-classical machine learning framework for the detection of AI-generated academic content. The system onboard the merging of Sentence-BERT semantic embeddings with quantum feature encoding that is supported by a 6-qubit, two-layer parameterized quantum circuit, thus resulting in a 390-dimensional hybrid representation which is classified through a lightweight multilayer perceptron. The framework has been tested on three benchmark datasets AI-GA, HWAI, and HAGT-1M to check its scalability and generalization to different academic writing situations. The results of the experiments show the framework has consistent and good discriminative capability achieving AUROC scores of more than 0.96 on all datasets and excellent performance (AUROC = 1.000, ACC = 99.98 %) on the large-scale data. Moreover, probability calibration by means of temperature scaling raises the trustworthiness of predicted confidence scores, leading to a 60 % reduction in the Expected Calibration Error (ECE) and without affecting the performance of the discrimination. When compared against the transformer-based and ensemble-learning detectors, HQML-NLP comes out with an equivalent and competitive detection accuracy and calibration quality yet demands more than 2000× less trainable parameters. These findings imply that hybrid quantum-classical representations act as an effective and compact alternative for the detection of AI-text in scholarly journals.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"191 ","pages":"Article 114634"},"PeriodicalIF":6.6,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146080200","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}

{"title":"Multiscale personalized federated load forecasting via enhanced ensemble empirical mode decomposition and Bayesian long short-term memory network","authors":"Yang Liu ,&nbsp;Yingchun Wang","doi":"10.1016/j.asoc.2026.114707","DOIUrl":"10.1016/j.asoc.2026.114707","url":null,"abstract":"<div><div>To enhance the personalized modeling capability of federated learning under non-independent and identically distributed (Non-IID) conditions, this study proposes a multiscale personalized federated framework for load forecasting. The framework employs a dual-model architecture that achieves “global trend sharing” and “local feature modeling,” effectively balancing collaboration and personalization across heterogeneous clients. At the data level, an enhanced ensemble empirical mode decomposition (E-EEMD) with paired noise cancellation and adaptive amplitude control is developed to suppress mode mixing and residual noise. A three-dimensional classification criterion integrating zero-crossing rate, Pearson correlation, and variance contribution enables adaptive component identification and multiscale feature extraction. At the model level, an enhanced Bayesian long short-term memory network (B-LSTM) incorporates temporal representation and uncertainty quantification, supporting confidence-weighted aggregation within federated learning. Experimental results across multiple regions verify that the proposed framework consistently outperforms conventional approaches in forecasting accuracy, robustness, and privacy preservation.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"191 ","pages":"Article 114707"},"PeriodicalIF":6.6,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146080199","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}

{"title":"Towards maritime safety with industrial multimodal models: Compliance reasoning from heterogeneous maritime data","authors":"Shen Wang ,&nbsp;Kejun Wu ,&nbsp;Renjie Qiao ,&nbsp;Runtian Wang ,&nbsp;Chengtao Cai","doi":"10.1016/j.asoc.2026.114701","DOIUrl":"10.1016/j.asoc.2026.114701","url":null,"abstract":"<div><div>The advancement of Industrial Large Models offers significant potential for enhancing intelligence and automation within specialized industrial sectors. This paper addresses the maritime industry, a critical domain where ensuring safety and regulatory compliance demands robust understanding and reasoning over complex, heterogeneous data types. Traditional methods often lack the necessary interpretability and generalization capabilities for dynamic maritime environments. To overcome these limitations, we introduce a novel framework leveraging an Industrial Large Model approach for behavioral compliance reasoning. Our methodology utilizes a transformer-based large multimodal model adept at processing diverse industrial maritime data, specifically remote sensing imagery (Synthetic Aperture Radar) and structured navigational records (Automatic Identification System). The system features two core contributions: (1) a large-scale dataset constructed with diverse maritime scenarios, integrating high-resolution visual data, structured metadata, and natural language descriptions tailored for industrial maritime contexts; and (2) a transformer-based vision-language model fine-tuned on this dataset to jointly perform vessel detection, identification, and behavioral compliance assessment—tasks crucial for industrial maritime operations. To improve the model’s reasoning depth and align with the requirements for trustworthy Artificial Intelligence in industrial settings, we incorporate Chain-of-Thought prompting, facilitating transparent, step-by-step decision-making processes. Extensive experiments validate our model’s high accuracy, interpretability, and generalization capabilities in complex operational settings, significantly enhancing anomaly detection and supporting reliable maritime safety decisions. This study underscores the capability of Industrial Large Models to interpret complex multimodal industrial data for enhanced maritime situational awareness, providing a practical illustration of adapting foundational models for safety-critical and regulatory compliance tasks, thereby advancing the understanding and application of large model technology within specialized industrial sectors.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"191 ","pages":"Article 114701"},"PeriodicalIF":6.6,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146080301","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}

{"title":"A generalized multi-agent deep reinforcement learning framework for enhancing resilience and energy management in smart grids","authors":"Pandia Rajan Jeyaraj,&nbsp;Siva Prakash Asokan,&nbsp;Aravind Chellachi Kathiresan,&nbsp;Edward Rajan Samuel Nadar","doi":"10.1016/j.asoc.2026.114793","DOIUrl":"10.1016/j.asoc.2026.114793","url":null,"abstract":"<div><div>With the integration of advanced computational algorithms and communication technologies, the smart grid functions as a Cyber-Physical System (CPS), making it vulnerable to cyber-attacks. Enhancing resilience is essential to mitigate these attacks, reduce outages, ensure secure operation, and maintain optimal power dispatch. This research proposes a cyber-constrained Multi-Agent Deep Reinforcement Learning (MADRL) framework for smart grids with distributed renewable energy sources, integrated with cyber-attack mitigation capabilities. The proposed MADRL algorithm enhances security and voltage stability by forming dynamic microgrids using data-driven Markov Decision Processes. It minimizes communication delays by avoiding congestion at non-cooperative nodes through constraint-based monitoring. To counter cyber intrusions, MADRL enables adaptive microgrid formation and learns to detect and respond to attacks through cyber-constrained training. The effectiveness of the proposed approach is validated on the IEEE 300-bus system under various outage and attack scenarios. Real-time simulations on IBM ILOG CPLEX optimization studio demonstrate its capability for online decision-making and resilience enhancement. Results confirm that the MADRL algorithm significantly improves smart grid reliability and security during cyber threats.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"192 ","pages":"Article 114793"},"PeriodicalIF":6.6,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146190703","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}

{"title":"A dual-strategy multi-objective evolutionary algorithm with domain-informed mutation and secondary weighted selection for counterfactual explanations in knowledge tracing","authors":"Lu Gao ,&nbsp;Xin Yuan ,&nbsp;Xiaoxuan Shen ,&nbsp;Jianwen Sun","doi":"10.1016/j.asoc.2026.114777","DOIUrl":"10.1016/j.asoc.2026.114777","url":null,"abstract":"<div><div>Deep learning-based knowledge tracing (DLKT) models enhance personalized learning with strong predictions. However, their black-box nature hinders educational use, raising concerns about transparency and trust. To address this issue, this paper proposes a dual-strategy multiobjective evolutionary algorithm (DSMOEA) for counterfactual explanation, aimed at providing effective post hoc interpretations of DLKT predictions. Leveraging causal reasoning, DSMOEA avoids spurious correlations and generates actionable counterfactuals by constructing a tutoring-oriented framework that yields optimal, concise practice sequences to enhance future learning gains. We formulate the counterfactual generation task as a multi-objective optimization problem and solve it with a MOEA. Two strategies improve efficiency and quality: domain-informed mutation (DIM) guides solution generation, and secondary weighted selection (SWS) ranks Pareto candidates by educational relevance. Experiments demonstrate superior effectiveness and sparsity over baselines, delivering reliable counterfactuals and actionable guidance for learning path optimization.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"192 ","pages":"Article 114777"},"PeriodicalIF":6.6,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146190707","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}

{"title":"Intuitionistic fuzzy-valued Choquet operator: theoretical foundations and application to aggregation tasks","authors":"Michał Boczek ,&nbsp;Marek Kaluszka ,&nbsp;Paweł Karczmarek ,&nbsp;Alicja Żmudzińska ,&nbsp;Albert Rachwał ,&nbsp;Michał Dolecki","doi":"10.1016/j.asoc.2026.114751","DOIUrl":"10.1016/j.asoc.2026.114751","url":null,"abstract":"<div><div>Aggregation function plays a fundamental role in decision-making problems involving incomplete information. In particular, the intuitionistic fuzzy-valued Choquet operator (IFC - operator), based on an Archimedean <span><math><mi>t</mi></math></span>-norm, provides a flexible and powerful framework for modeling interactions among criteria in intuitionistic fuzzy environments. Previous studies have mainly focused on examining whether the IFC - operator satisfies the axioms of an aggregation function for selected Archimedean <span><math><mi>t</mi></math></span>-norms. In the paper, we present a complete characterization that allows one to determine whether the IFC - operator qualifies as an aggregation function under a fixed total order. To achieve this, we establish a general theoretical framework that characterizes the conditions under which a generalized Choquet-like operator of the form <span><math><mrow><mtext>F</mtext></mrow><mo>(</mo><msub><mrow><mi>x</mi></mrow><mrow><mi>σ</mi><mo>(</mo><mn>1</mn><mo>)</mo></mrow></msub><mo>,</mo><mo>…</mo><mo>,</mo><msub><mrow><mi>x</mi></mrow><mrow><mi>σ</mi><mo>(</mo><mi>n</mi><mo>)</mo></mrow></msub><mo>,</mo><mi>μ</mi><mo>(</mo><msub><mi>B</mi><mrow><mi>σ</mi><mo>(</mo><mn>1</mn><mo>)</mo></mrow></msub><mo>)</mo><mo>,</mo><mo>…</mo><mo>,</mo><mi>μ</mi><mo>(</mo><msub><mi>B</mi><mrow><mi>σ</mi><mo>(</mo><mi>n</mi><mo>)</mo></mrow></msub><mo>)</mo><mo>)</mo></math></span> satisfies the axioms of aggregation functions given a fixed total order, where the values <span><math><msub><mrow><mi>x</mi></mrow><mrow><mi>σ</mi><mo>(</mo><mi>i</mi><mo>)</mo></mrow></msub></math></span> are taken from an arbitrary linearly ordered set and <span><math><mo>(</mo><msub><mi>B</mi><mrow><mi>σ</mi><mo>(</mo><mi>i</mi><mo>)</mo></mrow></msub><mo>)</mo></math></span> forms a monotonicity family of subsets of <span><math><mo>{</mo><mn>1</mn><mo>,</mo><mo>…</mo><mo>,</mo><mi>n</mi><mo>}</mo></math></span>. We also present some applications of the IFC - operator to combine the results of individual classifiers in the multi-class problems. We show that the IFC - operator can be effectively applied to the multi-class problems as the aggregation function to combine the results of individual classifiers. Extensive numerical experiments confirmed the effectiveness of the proposed method, which achieved high accuracy and consistently outperformed the Choquet integral as well as individual classifiers.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"192 ","pages":"Article 114751"},"PeriodicalIF":6.6,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146191837","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}

{"title":"A rate-distortion optimal evolutionary algorithm under the fixed bit-rate constraints for the JPEG quantization table optimization","authors":"Qijun Wang ,&nbsp;Hongyan Wei ,&nbsp;Ke Xu ,&nbsp;Bo Du ,&nbsp;Xingyi Zhang","doi":"10.1016/j.asoc.2026.114697","DOIUrl":"10.1016/j.asoc.2026.114697","url":null,"abstract":"<div><div>JPEG has been widely utilized across diverse applications and devices for approximately 30 years, and it will continue to dominate the image ecosystem for an extended duration. Optimizing the quantization table in JPEG compression not only maintains compatibility with the JPEG standard but also enhances compression efficiency, which is crucial for image storage and transmission. Existing Evolutionary Algorithm (EA) based optimization of JPEG quantization tables primarily focuses on optimizing the luminance quantization tables, but the optimization cannot be tailored for specified bit-rates. To address this problem, this paper proposes a rate-distortion optimal evolutionary algorithm with fixed bit-rates for the JPEG quantization table optimization, which optimizes the quantization tables through a co-evolutionary algorithm with two populations: one with bit-rate constraints and the other without any constraints, adopting the rate-distortion optimal principle based population update and different environmental selection strategies for each population. The two populations are evolved simultaneously and cooperate with each other to achieve better optimization performance. Experimental results on classical benchmark datasets demonstrate that our method can achieve the better quantization tables than current state-of-the-art algorithms in both compression efficiency and image quality while more accurate target bit-rate for JPEG compression can also be achieved through our method.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"192 ","pages":"Article 114697"},"PeriodicalIF":6.6,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146191835","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}

{"title":"ODPL: An objective-driven decomposed policy learning framework for urban autonomous driving control","authors":"Lizhi Bi ,&nbsp;Bin Shuai ,&nbsp;Zhanru Liu ,&nbsp;Yichen Sun ,&nbsp;Feiyu Yang","doi":"10.1016/j.asoc.2026.114781","DOIUrl":"10.1016/j.asoc.2026.114781","url":null,"abstract":"<div><div>In urban driving (UD) environments, deep reinforcement learning (DRL) demonstrates superior adaptability for autonomous driving control. However, fundamental challenges remain unsolved in existing DRL methods: error accumulation in Actor-Critic architectures, inconsistent learning signals from hand-crafted reward functions, and signal confusion in multi-dimensional control tasks. This paper proposes an Objective-Driven Decomposed Policy Learning (ODPL) framework for reliable urban autonomous driving control. The framework deconstructs the traditional reinforcement learning paradigm by eliminating Critic networks and directly utilizing driving-specific objectives as learning signals, thereby improving learning efficiency and accelerating convergence. ODPL employs a task decomposition strategy that decouples steering and throttle control into independent subtasks, resolving the signal confusion problem inherent in unified network architectures. Furthermore, to address credit assignment issues caused by physical delays, ODPL integrates a delay-aware state transition model that establishes precise causal relationships between current states and historical actions through a delay buffering mechanism. Simulation results in the CARLA environment demonstrate that the proposed ODPL framework achieves 70%–110% performance improvements over advanced baseline methods across various driving scenarios. The framework exhibits exceptional performance in complex curve scenarios while maintaining stable learning convergence under cold-start conditions without pre-training data.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"192 ","pages":"Article 114781"},"PeriodicalIF":6.6,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146191833","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}

{"title":"MTCCM: A multi-task conflict control model for Web API recommendation with semantic enhancement","authors":"Kezhou Chen ,&nbsp;Hongbin Zhang ,&nbsp;Zhengming Chen ,&nbsp;Lianglun Cheng ,&nbsp;Xu Lu","doi":"10.1016/j.asoc.2026.114749","DOIUrl":"10.1016/j.asoc.2026.114749","url":null,"abstract":"<div><div>With the rapid growth of web application programming interfaces (Web APIs), mashup development has emerged as a key paradigm that integrates existing APIs to meet the complex needs of various applications. The increasing number of Web APIs complicates selection and reduces development efficiency, while deep learning-based recommendation methods struggle with data sparsity and unstructured inputs. Multi-task learning (MTL) offers a promising solution by leveraging related tasks; however, many existing approaches fail to account for task conflicts, thereby limiting their effectiveness. We propose a multi-task conflict control model (MTCCM) for Web API recommendation to address these issues, integrating three synergistic modules. The semantic enhancement module (SEM) leverages a large language model with prompt-based inputs to structure unorganized mashups and Web API descriptions, enhancing semantic representation and mitigating issues with unstructured data. The conflict control module (CCM) operates in two stages: task conflict analysis identifies optimal auxiliary tasks, and gradient conflict control refines training by adjusting the direction of conflicting gradients. The neural network module (NNM) adopts a shared architecture with two task-specific outputs: one for Web API prediction and the other for Web API classification, which serves as an auxiliary task to improve the primary prediction performance. Extensive experiments demonstrate that MTCCM significantly outperforms state-of-the-art baselines, establishing a new benchmark in multi-task Web API recommendation.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"192 ","pages":"Article 114749"},"PeriodicalIF":6.6,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146191832","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}