Computer Science Review最新文献

{"title":"A survey of large language models for legal tasks: Progress, prospects and challenges","authors":"Congqing He ,&nbsp;Haichuan Hu ,&nbsp;Yanli Li ,&nbsp;Hao Zhang ,&nbsp;Quanjun Zhang","doi":"10.1016/j.cosrev.2026.100906","DOIUrl":"10.1016/j.cosrev.2026.100906","url":null,"abstract":"<div><div>Recent advances in large language models (LLMs) have unlocked new opportunities for machine learning and deep learning applications in the legal domain. LLMs demonstrate remarkable capabilities in comprehending complex legal language, analyzing lengthy documents, and generating contextually relevant legal text. This survey provides a task-oriented overview of the application of LLMs in the legal domain, focusing on the categorization of tasks and a review of associated methods, datasets and benchmarks. We first review both general and legal-domain LLMs, summarize their foundational architectures and adaptation methods, and briefly analyze their suitability for various legal tasks. Subsequently, we categorize LLM-based legal tasks along the typical legal workflow, including legal information retrieval, legal document analysis, judicial decision prediction, legal question answering, legal document generation, legal agent-based modeling, and legal education and training. Within each task, we primarily analyze specific methodologies such as retrieval-augmented generation, prompting strategies, and reasoning. Furthermore, a comprehensive collection of legal datasets, benchmarks, and model resources is presented as a practical reference for researchers and practitioners. Finally, we outline open challenges and future directions, addressing issues such as bias, interpretability, data privacy, and regulatory compliance in legal LLMs. This survey provides a structured and comprehensive overview to facilitate the adoption and further development of LLMs in the legal domain. The collection is available at <span><span>https://github.com/hecongqing/Awesome-LLM4Law</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":48633,"journal":{"name":"Computer Science Review","volume":"60 ","pages":"Article 100906"},"PeriodicalIF":12.7,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146014532","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":"Three-dimensional visualization of X-ray micro-CT with large-scale datasets: Efficiency and accuracy for real-time interaction","authors":"Yipeng Yin ,&nbsp;Rao Yao ,&nbsp;Qingying Li ,&nbsp;Dazhong Wang ,&nbsp;Hong Zhou ,&nbsp;Zhijun Fang ,&nbsp;Jianing Chen ,&nbsp;Longjie Qian ,&nbsp;Mingyue Wu","doi":"10.1016/j.cosrev.2025.100888","DOIUrl":"10.1016/j.cosrev.2025.100888","url":null,"abstract":"<div><div>As Micro-CT technology continues to refine its characterization of material microstructures, industrial CT ultra-precision inspection is generating increasingly large datasets, necessitating solutions to the trade-off between accuracy and efficiency in the 3D characterization of defects during ultra-precise detection. This article provides a unique perspective on recent advances in accurate and efficient 3D visualization using Micro-CT, tracing its evolution from medical imaging to industrial non-destructive testing (NDT). Among the numerous CT reconstruction and volume rendering methods, this article selectively reviews and analyzes approaches that balance accuracy and efficiency, offering a comprehensive analysis to help researchers quickly grasp highly efficient and accurate 3D reconstruction methods for microscopic features. By comparing the principles of computed tomography with advancements in microstructural technology, this article examines the evolution of CT reconstruction algorithms from analytical methods to deep learning techniques, as well as improvements in volume rendering algorithms, acceleration, and data reduction. Additionally, it explores advanced lighting models for high-accuracy, photorealistic, and efficient volume rendering. Furthermore, this article envisions potential directions in CT reconstruction and volume rendering. It aims to guide future research in quickly selecting efficient and precise methods and developing new ideas and approaches for real-time online monitoring of internal material defects through virtual-physical interaction, for applying digital twin model to structural health monitoring (SHM).</div></div>","PeriodicalId":48633,"journal":{"name":"Computer Science Review","volume":"60 ","pages":"Article 100888"},"PeriodicalIF":12.7,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145884463","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 review of self-scheduled strategies for numerical linear algebra on GPU","authors":"Manuel Freire,&nbsp;Ernesto Dufrechou,&nbsp;Pablo Ezzatti","doi":"10.1016/j.cosrev.2026.100901","DOIUrl":"10.1016/j.cosrev.2026.100901","url":null,"abstract":"<div><div>The slowdown of Moore’s Law has generated a shift toward multicore and massively parallel architectures, making it essential to efficiently exploit parallelism for high performance. Among task-scheduling paradigms for shared-memory systems, self-scheduled (or sync-free) methods dynamically assign tasks to processors, balancing workload and improving utilization while managing data dependencies through synchronization mechanisms. Originally applied to sparse linear algebra on CPUs, these techniques have recently been adapted to GPUs. The introduction of the self-scheduled approach for sparse triangular solves (<span><math><mi>S</mi><mi>p</mi><mi>T</mi><mi>R</mi><mi>S</mi><mi>V</mi></math></span>) marked a breakthrough, outperforming static level-set methods and motivating further work. While the first efforts related to this paradigm were focused on the <span><math><mi>S</mi><mi>p</mi><mi>T</mi><mi>R</mi><mi>S</mi><mi>V</mi></math></span> operation, the efficiency of this strategy has extended to multiple Numerical Linear Algebra (NLA) kernels. This paper surveys the state of the art in self-scheduled algorithms for numerical linear algebra on GPUs. We classify contributions into three categories: (i) optimizations of triangular solvers, (ii) preprocessing techniques that enhance sync-free execution, and (iii) extensions of the paradigm to other linear algebra kernels.</div></div>","PeriodicalId":48633,"journal":{"name":"Computer Science Review","volume":"60 ","pages":"Article 100901"},"PeriodicalIF":12.7,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146072496","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":"Exploring transparency in pathological image analysis: A comprehensive review of explainable artificial intelligence (XAI) techniques","authors":"Lingling Yuan ,&nbsp;Yutong Gu ,&nbsp;Yan Han ,&nbsp;Tianming Du ,&nbsp;Marcin Grzegorzek ,&nbsp;Chen Li","doi":"10.1016/j.cosrev.2025.100863","DOIUrl":"10.1016/j.cosrev.2025.100863","url":null,"abstract":"<div><div>Pathological image analysis is a key component of digital pathology. The integration of artificial intelligence (AI) has substantially enhanced diagnostic efficiency. However, these models often lack transparency, and their black-box nature limits clinical trust and broader adoption. Although existing reviews have explored the application of explainable AI (XAI) in pathology, most use incomplete or inconsistent taxonomies and lack evaluation frameworks that consider multiple dimensions. The lack of a clear taxonomy and evaluation framework makes it difficult to compare results, identify suitable approaches, and assess the strengths and limitations of XAI methods from multiple perspectives. To address these gaps, this review presents a taxonomy from five perspectives and introduces the PathXAI-2 L minimal evaluation framework. Specifically, it provides a comprehensive overview of key tasks in pathological image analysis, commonly used datasets and their biases, and preprocessing strategies, algorithmic frameworks, and evaluation metrics. Based on this foundation, it proposes a five-dimensional taxonomy of XAI methods, organized by model dependency, explanation scope, explanation stage, explanatory modality, and evidential strength. The evidence perspective is used to distinguish between correlation and causality, by grouping explanations into association, sensitivity, and intervention levels. This review further examines two important factors influencing explainability results: the sensitivity of XAI methods to dataset choice, and the role of external resources in supporting explanations. At the technical level, this review groups XAI methods into seven representative types and introduces XAI tools in pathology. Finally, it proposes a minimal evaluation framework called PathXAI-2 L, focused on the functional and human layers, while highlighting key challenges and future directions for application-grounded evaluation. By integrating tasks, datasets, methods, and evaluation strategies, this review provides a comprehensive reference for advancing both technical progress and clinical impact in pathology XAI.</div></div>","PeriodicalId":48633,"journal":{"name":"Computer Science Review","volume":"60 ","pages":"Article 100863"},"PeriodicalIF":12.7,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145610126","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":"Recommender systems and sustainability: a dual perspective","authors":"Allegra De Filippo ,&nbsp;Giuseppe Spillo ,&nbsp;Ludovico Boratto ,&nbsp;Michela Milano ,&nbsp;Cataldo Musto ,&nbsp;Giovanni Semeraro","doi":"10.1016/j.cosrev.2026.100912","DOIUrl":"10.1016/j.cosrev.2026.100912","url":null,"abstract":"<div><div>The concept of sustainability, as outlined by the United Nations’ Sustainable Development Goals (SDGs), refers to the ability to meet the needs of the present without compromising the ability of future generations to meet their own needs. This vision is addressed by combining goals concerning the <em>environmental</em>, <em>social</em>, and <em>economic</em> spheres. In this context, Recommender Systems (RS) have emerged as tools that can foster these principles by nudging responsible user behavior and promoting sustainable decision-making. However, the interplay between RS and sustainability is inherently complex since it can be analyzed from two different perspectives: <em>(i) RS for Sustainability</em>, which focuses on how recommendation algorithms can support the achievement of SDGs, and <em>(ii) Sustainability of RS</em>, which focuses on developing recommendation models that inherently adhere to sustainability principles. While the integration of both these perspectives is beneficial and crucial, unfortunately, the current literature has addressed these aspects independently. Accordingly, in this survey, we first provide a comprehensive review of the existing literature on RS that either promotes sustainable behaviors aligned with the SDGs or embeds sustainability principles into their algorithmic design. Next, we identify current gaps and propose key research directions toward an integrated, holistic approach that concurrently addresses both aspects to advance the development of sustainable RS.</div></div>","PeriodicalId":48633,"journal":{"name":"Computer Science Review","volume":"60 ","pages":"Article 100912"},"PeriodicalIF":12.7,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146033306","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":"Backdoors to satisfaction continued","authors":"Serge Gaspers ,&nbsp;Stefan Szeider","doi":"10.1016/j.cosrev.2025.100868","DOIUrl":"10.1016/j.cosrev.2025.100868","url":null,"abstract":"<div><div>A backdoor set is a set of variables of a propositional formula such that fixing the truth values of the variables in the backdoor set moves the formula into some polynomial-time decidable class. If we know a small backdoor set, we can reduce the question of whether the given formula is satisfiable to the same question for one or several easy formulas that belong to the tractable class under consideration. Continuing our 2012 survey, we review parameterized complexity results for problems that arise in the context of backdoor sets, such as the problem of finding a backdoor set of size, depth, or treewidth at most <span><math><mi>k</mi></math></span>, parameterized by <span><math><mi>k</mi></math></span>.</div></div>","PeriodicalId":48633,"journal":{"name":"Computer Science Review","volume":"60 ","pages":"Article 100868"},"PeriodicalIF":12.7,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145732497","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":"Advanced computational models for urban traffic flow prediction: A comprehensive review and future directions","authors":"Ahmad Ali ,&nbsp;Amin Sharafian ,&nbsp;H.M. Yasir Naeem ,&nbsp;Muhammad Zakarya ,&nbsp;Zongze Wu ,&nbsp;Xiaoshan Bai","doi":"10.1016/j.cosrev.2025.100886","DOIUrl":"10.1016/j.cosrev.2025.100886","url":null,"abstract":"<div><div>Traffic flow prediction is a fundamental task in intelligent transportation systems (ITS), supporting efficient mobility management and smart city development. In recent years, ITS research has rapidly progressed from traditional statistical models to advanced deep learning architectures, including convolutional, recurrent, graph-based, and attention-driven spatio-temporal networks. This article provides a comprehensive review of these approaches, categorizing them by methodological families, summarizing their strengths and limitations, and comparing their performance on widely used benchmarks. A particular emphasis is placed on federated learning, an emerging paradigm that enables collaborative model training across cities, operators, and edge devices without exposing sensitive data. We outline key application scenarios for federated traffic prediction, analyze technical challenges such as independent and identically distributed (IID) and non-IID data distributions, communication overheads, and privacy risks, and highlight representative solutions proposed in the recent literature. In addition, we compile a repository of publicly available datasets and summarize benchmark results to facilitate reproducibility and fair comparison. Finally, we identify open challenges and promising directions, including federated graph learning, explainable and trustworthy AI, and resource-aware deployment. This review aims to serve as a reference for researchers and practitioners, offering both a structured overview of the state-of-the-art and a roadmap for future advances in traffic flow prediction.</div></div>","PeriodicalId":48633,"journal":{"name":"Computer Science Review","volume":"60 ","pages":"Article 100886"},"PeriodicalIF":12.7,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145823148","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":"Fusing LLMs and diffusion models: A comprehensive survey of progress, challenges, and future directions in generative AI","authors":"Bilel Benjdira ,&nbsp;Anas M. Ali ,&nbsp;Wadii Boulila ,&nbsp;Anis Koubaa","doi":"10.1016/j.cosrev.2025.100881","DOIUrl":"10.1016/j.cosrev.2025.100881","url":null,"abstract":"<div><div>Despite the rapid advancements in generative AI, the integration of diffusion models and Large Language Models (LLMs) remains largely underexplored, with only a few studies systematically addressing this research frontier. This emerging research direction holds significant potential for advancing multimodal generation, reasoning, and cross-domain understanding through the complementary strengths of both model families. This survey addresses this critical gap by providing a comprehensive analysis of recent progress in LLM–diffusion model integration. Methodologically, the study examines approaches such as latent space alignment, prompt engineering, and novel architectures that facilitate synergy between LLMs and diffusion models. Key findings reveal that while this integration enhances generative capabilities, it also introduces challenges, including high computational costs, misalignment between modalities, data scarcity, and quality control issues. The survey systematically evaluates existing solutions to these challenges, highlighting their strengths, limitations, and practical implications. Emerging trends, such as efficient fine-tuning strategies, hybrid architectures, and multimodal data augmentation, are identified as promising avenues for future research. By synthesizing current knowledge and offering actionable insights, this survey serves as a valuable resource for researchers and practitioners seeking to explore the combined potential of LLMs and diffusion models. The repository for this survey is publicly available at <span><span>https://github.com/AnasHXH/Connecting-LLMs-to-Diffusion-Models-A-Survey</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":48633,"journal":{"name":"Computer Science Review","volume":"60 ","pages":"Article 100881"},"PeriodicalIF":12.7,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145823149","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 systematic review of deep learning-based models for elderly and human activity recognition","authors":"Mukesh Dalal ,&nbsp;Payal Mittal","doi":"10.1016/j.cosrev.2025.100879","DOIUrl":"10.1016/j.cosrev.2025.100879","url":null,"abstract":"<div><div>The ageing population in India and globally has highlighted the significant need for research on elderly activity detection and enhancing safety measures for elderly individuals living alone. This is crucial due to their increased vulnerability to accidents when unattended. By monitoring their behaviors and providing prompt assistance, potential dangers can be mitigated. Recently, deep learning and sensor-based methods have been applied in activity recognition; however, Elderly Activity Recognition (EAR) research is limited compared to Human Activity Recognition (HAR). Existing surveys are failing to provide a detailed analysis of datasets, models, and performance metrics. This paper presents a systematic and detailed review of existing state-of-the-art deep learning models for EAR and HAR based on the PRISMA approach. Further, a comparative analysis of the existing literature is done by using standard evaluation metrics. The paper delineates a wide range of publicly available datasets encompassing diverse subjects, numbers of instances, sensor or image/video modalities, and their respective download sources. The paper emphasises the research trends and highlights research gaps for EAR, including fall detection and reduced mobility patterns. It proposes adaptations of existing HAR methods to address these challenges and provides a practical guide for choosing suitable models and datasets for monitoring elderly activities, thus filling a gap in current research.</div></div>","PeriodicalId":48633,"journal":{"name":"Computer Science Review","volume":"60 ","pages":"Article 100879"},"PeriodicalIF":12.7,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145731224","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":"DDoS attack detection and defense techniques in software defined networks: A survey","authors":"Wei Wang ,&nbsp;Yong Liu ,&nbsp;Qian Meng ,&nbsp;Zihang Chen","doi":"10.1016/j.cosrev.2026.100921","DOIUrl":"10.1016/j.cosrev.2026.100921","url":null,"abstract":"<div><div>Traditional network architectures face operational rigidity due to static device configurations, which constrain network management efficiency and functional evolution. Software-Defined Networking (SDN) overcomes these limitations through controller-based centralized management, significantly improving network programmability. However, the fundamental separation between control and data planes introduces critical vulnerabilities to distributed denial-of-service (DDoS) attacks, where attackers can disable SDN operations by exhausting controller resources or saturating control channels. Recent research has produced various innovative solutions to address SDN-specific DDoS threats. This paper systematically reviews advancements in DDoS defense mechanisms for SDN environments, including the development of attack taxonomies with characteristic features, comprehensive assessments of architectural vulnerabilities, and critical evaluations of detection methodologies. Finally, we analyze the limitations of existing countermeasures against DDoS attacks and discussing potential development directions for future defense strategies.</div></div>","PeriodicalId":48633,"journal":{"name":"Computer Science Review","volume":"60 ","pages":"Article 100921"},"PeriodicalIF":12.7,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146095848","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}