Intelligent Systems with Applications最新文献

{"title":"Integrated framework of fragment-based method and generative model for lead drug molecules discovery","authors":"Uche A.K. Chude-Okonkwo,&nbsp;Odifentse Lehasa","doi":"10.1016/j.iswa.2025.200508","DOIUrl":"10.1016/j.iswa.2025.200508","url":null,"abstract":"<div><div>Generative models have proven valuable in generating novel lead molecules with drug-like properties. However, beyond generating drug-like molecules, the generative model should also be able to create drug molecules with structural properties and pharmacophores to modulate a specific disease. The molecular generation process should also address the multi-objective optimization challenge of producing molecules with the desired efficacy and minimal side effects. This may entail the generation of a diverse pool of molecules with the desired structural properties and pharmacophore, which would offer diverse options and paths to developing potential new drug candidates by prioritizing molecules that balance the desired properties that can cater to the needs of different individuals. Achieving this requires a generative model learning a large dataset of molecular instances with the desired chemical/structural properties. However, large sets of drug molecules are not readily available for many diseases as there are few known drug molecular instances for treating any disease. To address this challenge, this paper presents an <em>in silico</em> molecular generative framework aided by fragment-based molecules’ synthesis for generating a pool of lead molecular instances possessing structural properties and pharmacophores to treat a disease of interest. The operation of the framework is explored using Hypertension as the disease of interest and beta-blocker as the reference hypertension drug to be generated. We generated over 123 beta-blocker-like molecules and further virtual-screened them for drug-likeness, docking probability, scaffold diversity, electrostatic complementarity, and synthesis accessibility to arrive at the final lead beta-blocker-like molecules.</div></div>","PeriodicalId":100684,"journal":{"name":"Intelligent Systems with Applications","volume":"26 ","pages":"Article 200508"},"PeriodicalIF":0.0,"publicationDate":"2025-03-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143715269","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Deep learning for efficient high-resolution image processing: A systematic review","authors":"Albert Dede ,&nbsp;Henry Nunoo-Mensah ,&nbsp;Eric Tutu Tchao ,&nbsp;Andrew Selasi Agbemenu ,&nbsp;Prince Ebenezer Adjei ,&nbsp;Francisca Adoma Acheampong ,&nbsp;Jerry John Kponyo","doi":"10.1016/j.iswa.2025.200505","DOIUrl":"10.1016/j.iswa.2025.200505","url":null,"abstract":"<div><div>High-resolution images are increasingly used in fields such as remote sensing, medical imaging, and agriculture, but they present significant computational challenges when processed with deep learning models. This paper provides a systematic review of deep learning techniques developed to improve the efficiency of high-resolution image processing. We investigate techniques like lightweight neural networks, vision transformers adapted for high-resolution inputs, and models using frequency-domain inputs based on 96 studies from 2018 to 2023. These techniques have many applications, from environmental monitoring and urban planning to disease diagnosis. We emphasize the advancements in efficient high-resolution deep learning models, discussing their performance in terms of accuracy, speed, and resource requirements. Key challenges, including the trade-off between processing efficiency and model accuracy, are analysed, and potential future research directions are proposed to address these issues.</div></div>","PeriodicalId":100684,"journal":{"name":"Intelligent Systems with Applications","volume":"26 ","pages":"Article 200505"},"PeriodicalIF":0.0,"publicationDate":"2025-03-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143705634","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Leveraging social networks as an optimization approach","authors":"Hamed Ghadirian ,&nbsp;Seyed Jalaleddin Mousavirad","doi":"10.1016/j.iswa.2025.200506","DOIUrl":"10.1016/j.iswa.2025.200506","url":null,"abstract":"<div><div>Metaheuristic algorithms have become powerful tools for solving complex optimization problems. Consensus-based optimization (CBO), inspired by social interactions, models a network where agents adjust their positions by learning from their neighbors. While effective, CBO relies on a fixed network structure, limiting its adaptability. To overcome this, we propose the Human Generation (HG) algorithm, which extends CBO by incorporating a two-layer influence mechanism. The first layer mimics kinship-based learning, ensuring local refinement, while the second layer models elite-following behavior, enabling efficient global exploration. This structured adaptation enhances both convergence speed and solution accuracy. We evaluate HG across unimodal, multimodal, and complex optimization problems, as well as a real-world image thresholding application. Experimental results demonstrate that HG consistently outperforms CBO and other state-of-the-art algorithms, making it a robust optimization approach.</div></div>","PeriodicalId":100684,"journal":{"name":"Intelligent Systems with Applications","volume":"26 ","pages":"Article 200506"},"PeriodicalIF":0.0,"publicationDate":"2025-03-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143704010","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A systematic literature review of quantum object detection and recognition: research trend, datasets, topics and methods","authors":"Ifran Lindu Mahargya,&nbsp;Guruh Fajar Shidik,&nbsp;Affandy,&nbsp;Pujiono,&nbsp;Supriadi Rustad","doi":"10.1016/j.iswa.2025.200499","DOIUrl":"10.1016/j.iswa.2025.200499","url":null,"abstract":"<div><div>Quantum computing is a computational process that utilizes quantum mechanics features, namely superposition, interference, and entanglement, in information processing, allowing computation to run in parallel. The advantage of quantum computing is that it solves complex problems whereas classical computing is impossible because it requires expensive computing costs. Object detection and recognition is a task of computer vision, where research in this field aims to improve the ability of computer algorithms to produce interpretations of visual information. Humans easily analyze and describe the visual information received. However, unlike computer systems, they must learn and explore using machine learning from the visual information received to provide correct interpretations of visual information. This paper presents a systematic review of papers published from 2012 to 2024 to answer how far quantum object detection and recognition research has been conducted. The methodology of this review follows a systematic literature review such as the method proposed by Kitchenham et al. The selected primary studies amounted to 29 papers from four source digital libraries. The application of quantum algorithms is more often used to improve the performance of classical computing. The quantum model category consists of 3 types, namely pure quantum, hybrid classical-quantum, and quantum-inspired ML. Hybrid classical-quantum is the most discussed model and Quantum Convolutional Neural Network is the most frequently discussed algorithm or model in image classification from 2012 to 2024. Quantum algorithms show good results and can improve the performance of classical algorithms, although currently, the ability of quantum computing is not fully optimal because the development of quantum computers is still in the noisy intermediate-scale quantum era. However, with the current limited quantum computing capabilities, it can already outperform the capabilities of classical computing. Based on this, studies on quantum object detection and recognition need to be carried out so that when the full potential of quantum computing can be utilized, the user's capacity is competent.</div></div>","PeriodicalId":100684,"journal":{"name":"Intelligent Systems with Applications","volume":"26 ","pages":"Article 200499"},"PeriodicalIF":0.0,"publicationDate":"2025-03-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143637305","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"An effective approach for early fuel leakage detection with enhanced explainability","authors":"Ruimin Chu ,&nbsp;Li Chik ,&nbsp;Yiliao Song ,&nbsp;Jeffrey Chan ,&nbsp;Xiaodong Li","doi":"10.1016/j.iswa.2025.200504","DOIUrl":"10.1016/j.iswa.2025.200504","url":null,"abstract":"<div><div>Leakage detection at service stations with underground storage tanks containing hazardous products, such as fuel, is a critical task. Early detection is important to halt the spread of leaks, which can pose significant economic and ecological impacts on the surrounding community. Existing fuel leakage detection methods typically rely on statistical analysis of low-granularity inventory data, leading to delayed detection. Moreover, explainability, a crucial factor for practitioners to validate detection outcomes, remains unexplored in this domain. To address these limitations, we propose an <strong>EX</strong>plainable <strong>F</strong>uel <strong>L</strong>eakage <strong>D</strong>etection approach called EXFLD, which performs online fuel leakage detection and provides intuitive explanations for detection validation. EXFLD incorporates a high-performance deep learning model for accurate online fuel leakage detection and an inherently interpretable model to generate intuitive textual explanations to assist practitioners in result validation. Unlike existing explainable artificial intelligence methods that often use deep learning models which can be hard to interpret, EXFLD is a human-centric system designed to provide clear and understandable insights to support decision-making. Through case studies, we demonstrate that EXFLD can provide intuitive and meaningful textual explanations that humans can easily understand. Additionally, we show that incorporating semantic constraints during training in the ANFIS model enhances the semantic interpretability of these explanations by improving the coverage and distinguishability of membership functions. Experimental evaluations using a dataset collected from real-world sites in Australia, encompassing 167 tank instances, demonstrate that EXFLD achieves competitive performance compared to baseline methods, with an F2-score of 0.7969. This dual focus on accuracy and human-centric explainability marks a significant advancement in fuel leakage detection, potentially facilitating broader adoption.</div></div>","PeriodicalId":100684,"journal":{"name":"Intelligent Systems with Applications","volume":"26 ","pages":"Article 200504"},"PeriodicalIF":0.0,"publicationDate":"2025-03-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143680449","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Learning how to transfer: A lifelong domain knowledge distillation framework for continual MRC","authors":"Songze Li ,&nbsp;Zhijing Wu ,&nbsp;Runmin Cao ,&nbsp;Xiaohan Zhang ,&nbsp;Yifan Wang ,&nbsp;Hua Xu ,&nbsp;Kai Gao","doi":"10.1016/j.iswa.2025.200497","DOIUrl":"10.1016/j.iswa.2025.200497","url":null,"abstract":"<div><div>Machine Reading Comprehension (MRC) has attracted wide attention in recent years. It can reflect how well a machine understands human language. Benefitting from the increasing large-scale benchmark and pre-trained language models, a lot of MRC models have achieved remarkable success and even exceeded human performance. However, real-world MRC systems need incrementally learn from a continuous data stream across time without accessing the previously seen data, called Continual MRC system. It is a great challenge to learn a new domain incrementally without catastrophically forgetting previous knowledge. In this paper, MK-MRC (an extension of MA-MRC), a continual MRC framework with uncertainty-aware fixed <strong>M</strong>emory and lifelong domain <strong>K</strong>nowledge distillation, is proposed. MK-MRC is a memory replaying based method, in which a fixed-size memory buffer stores a small number of samples in previous domain data along with an uncertainty-aware updating strategy when new domain data arrives. For incremental learning, MK-MRC fully uses the domain adaptation and transfer relationship between memory and new domain data through several domain knowledge distillation strategies.</div><div>Compared with MA-MRC, MK-MRC additionally introduces more strategies to strengthen the ability of continual learning, such as data augmentation and special task-related knowledge distillation. Experimental results show that MK-MRC yields consistent improvement compared with strong baselines and has a substantial incremental learning ability without catastrophically forgetting under four continual span-extractive and multiple-choice MRC settings.</div></div>","PeriodicalId":100684,"journal":{"name":"Intelligent Systems with Applications","volume":"26 ","pages":"Article 200497"},"PeriodicalIF":0.0,"publicationDate":"2025-03-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143619117","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Minimum cost of job assignment in polynomial time by adaptive unbiased filtering and branch-and-bound algorithm with the best predictor","authors":"Jeeraporn Werapun,&nbsp;Witchaya Towongpaichayont,&nbsp;Anantaporn Hanskunatai","doi":"10.1016/j.iswa.2025.200502","DOIUrl":"10.1016/j.iswa.2025.200502","url":null,"abstract":"<div><div>The minimum cost of job assignment (Min-JA) is one of the practical NP-hard problems to manage the optimization in science-and-engineering applications. Formally, the optimal solution of the Min-JA can be computed by the branch-and-bound (BnB) algorithm (with the efficient predictor) in O(<em>n</em>!), <em>n</em> = problem size, and O(<em>n</em>³) in the best case but that best case hardly occurs. Currently, metaheuristic algorithms, such as genetic algorithms (GA) and swarm-optimization algorithms, are extensively studied, for polynomial-time solutions. Recently, unbiased filtering (in search-space reduction) could solve some NP-hard problems, such as 0/1-knapsack and multiple 0/1-knapsacks with Latin square (LS) of m-capacity ranking, for the ideal solutions in polynomial time. To solve the Min-JA problem, we propose the adaptive unbiased-filtering (AU-filtering) in O(<em>n</em>³) with a new hybrid (search-space) reduction (of the indirect metaheuristic strategy and the exact BnB). Innovation-and-contribution of our AU-filtering is achieved through three main steps: 1. find 9 + <em>n</em> effective job-orders for the good initial solutions (by the indirect assignment with UP: unbiased predictor), 2. improve top 9-solutions by the indirect improvement of the significant job-orders (by Latin square of <em>n</em> permutations plus <em>n</em> complex mod-functions), and 3. classify objects (from three of the best solutions) for AU-filtering (on large <em>n</em>) with deep-reduction (on smaller <em>n</em>’) and repeat (1)-(3) until <em>n</em>’ &lt; 6, the exact BnB is applied. In experiments, the proposed AU-filtering was evaluated by a simulation study, where its ideal results outperformed the best results of the hybrid swarm-GA algorithm on a variety of 2D datasets (<em>n</em> ≤ 1000).</div></div>","PeriodicalId":100684,"journal":{"name":"Intelligent Systems with Applications","volume":"26 ","pages":"Article 200502"},"PeriodicalIF":0.0,"publicationDate":"2025-03-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143580565","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A unified prompt-based framework for few-shot multimodal language analysis","authors":"Xiaohan Zhang ,&nbsp;Runmin Cao ,&nbsp;Yifan Wang ,&nbsp;Songze Li ,&nbsp;Hua Xu ,&nbsp;Kai Gao ,&nbsp;Lunsong Huang","doi":"10.1016/j.iswa.2025.200498","DOIUrl":"10.1016/j.iswa.2025.200498","url":null,"abstract":"<div><div>Multimodal language analysis is a trending topic in NLP. It relies on large-scale annotated data, which is scarce due to its time-consuming and labor-intensive nature. Multimodal prompt learning has shown promise in low-resource scenarios. However, previous works either cannot handle semantically complex tasks, or involve too few modalities. In addition, most of them only focus on prompting language modality, disregarding the untapped potential of other modalities. We propose a unified prompt-based framework for few-shot multimodal language analysis . Specifically, based on pretrained language model, our model can handle semantically complex tasks involving text, audio and video modalities. To enable more effective utilization of video and audio modalities by the language model, we introduce semantic alignment pre-training to bridge the semantic gap between them and the language model, alongside effective fusion method for video and audio modalities. Additionally, we introduce a novel effective prompt method — Multimodal Prompt Encoder — to prompt the entirety of multimodal information. Extensive experiments conducted on six datasets under four multimodal language subtasks demonstrate the effectiveness of our approach.</div></div>","PeriodicalId":100684,"journal":{"name":"Intelligent Systems with Applications","volume":"26 ","pages":"Article 200498"},"PeriodicalIF":0.0,"publicationDate":"2025-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143549364","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A comprehensive fusion model for improved pneumonia prediction based on KNN-wavelet-GLCM and a residual network","authors":"Asmaa Shati ,&nbsp;Ghulam Mubashar Hassan ,&nbsp;Amitava Datta","doi":"10.1016/j.iswa.2025.200492","DOIUrl":"10.1016/j.iswa.2025.200492","url":null,"abstract":"<div><div>Pneumonia is a severe disease that contributes to global mortality rates, emphasizing the critical need for early detection to improve patient survival. Chest radiography (X-ray) images serve as a fundamental diagnostic tool in clinical practice to detect various lung abnormalities. However, medical images, particularly X-rays, contain crucial data that are often imperceptible to the human eye. This study presents a novel fusion model (Res-WG-KNN) based on a soft voting ensemble strategy to predict pneumonia from chest X-ray images. It utilizes 2D-discrete wavelet decomposition and texture features from the Gray Level Co-occurrence Matrix (GLCM) with supervised machine learning, alongside raw X-ray images using a modified Residual Network ResNet-50. The proposed model was evaluated using two public pneumonia X-ray image datasets: one for adult patients, called the Radiological Society of North America (RSNA) dataset, and one for pediatric patients, called the Kermany dataset. These datasets differ in both size and image format, with the RSNA dataset using DICOM images and the Kermany dataset using JPEG images. The use of a soft voting technique in the proposed model effectively enhances classification performance beyond current benchmarks, achieving 97.0% accuracy and 0.97 AUC on the RSNA dataset, and 99.0% accuracy with 0.99 AUC on the Kermany dataset for pneumonia prediction.</div></div>","PeriodicalId":100684,"journal":{"name":"Intelligent Systems with Applications","volume":"26 ","pages":"Article 200492"},"PeriodicalIF":0.0,"publicationDate":"2025-02-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143529748","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Emulating fundamental analysts: Analytical stage-based multi-agent framework enhanced with expert guidance and Preference-Anchored Likelihood Adjustment","authors":"Tao Xu ,&nbsp;Zhe Piao ,&nbsp;Tadashi Mukai ,&nbsp;Yuri Murayama ,&nbsp;Kiyoshi Izumi","doi":"10.1016/j.iswa.2025.200496","DOIUrl":"10.1016/j.iswa.2025.200496","url":null,"abstract":"<div><div>With the rapid advancement of large language models (LLMs), some studies have explored their potential for predicting stock prices based on financial texts. However, previous research often overlooked the depth of analysis generated by LLMs, resulting in reasoning processes inferior to those of human analysts. In fundamental investing, which requires in-depth company analysis, conclusions from imperfect reasoning lack persuasiveness. In this study, inspired by the analysis process of human analysts, we propose an “Analytical Stage-Based Multi-Agent Framework” to enable LLMs to perform in-depth fundamental analysis. This framework divides the analysis into multiple stages, assigning an LLM agent to each. We enhance each agent’s capabilities for its specific task through expert guidance or fine-tuning, allowing them to collectively emulate the workflow of human analysts. Furthermore, we introduce Preference-Anchored Likelihood Adjustment, a new method for fine-tuning LLMs. This approach addresses the decline in likelihood of generating correct responses that occurs after using existing preference alignment methods. It employs an objective function with two terms: one to increase likelihood and another to preserve aligned preference. We conducted experiments using our framework to analyze company earnings releases. We evaluated the analysis quality based on comprehensiveness and logical soundness, while correctness was assessed by using stock prices as the ground truth to calculate the Matthews correlation coefficient and F1 score. Results demonstrate that even without expert guidance and fine-tuning, our multi-agent framework can enhance LLMs in both analysis quality and correctness. When combined with expert guidance and fine-tuning, the performance is further improved.</div></div>","PeriodicalId":100684,"journal":{"name":"Intelligent Systems with Applications","volume":"26 ","pages":"Article 200496"},"PeriodicalIF":0.0,"publicationDate":"2025-02-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143529747","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}