Sumair Aziz, Girija Chetty, Roland Goecke, Raul Fernandez-Rojas
Healthcare Embedded Systems (HES) use biosensors to capture physiological data, analyse it with advanced algorithms, and provide timely alerts during emergencies. These systems enhance healthcare delivery by supporting diagnosis, early symptom detection, and disease prediction. Despite extensive research on data analysis techniques in healthcare, selecting real-time methods for specific embedded hardware remains challenging. This review aims to summarise and synthesise existing literature to: (a) identify the healthcare challenges addressed by HES and the types of biosignals employed, (b) explore the embedded platforms utilised for implementing HES, and (c) examine the data analysis techniques used for real-time HES applications. A systematic search across three electronic databases (2015-2024), identified 50 relevant studies. These studies span various application domains, biosensing modalities, feature extraction methods, and machine learning and deep learning techniques. Raspberry Pi single-board computers emerged as the most popular embedded platform for implementing AI-based HES. Deep learning, especially convolutional neural networks, dominated, with cardiac health as the primary focus. While the reviewed studies demonstrate promising results, they are often constrained by specific experimental contexts. This review offers a comprehensive overview of real-time data analysis in HES and highlights key opportunities for future research to advance the field.
{"title":"A Systematic Literature Review of Healthcare Embedded Systems Using AI-based Biosignal Analysis","authors":"Sumair Aziz, Girija Chetty, Roland Goecke, Raul Fernandez-Rojas","doi":"10.1145/3793669","DOIUrl":"https://doi.org/10.1145/3793669","url":null,"abstract":"Healthcare Embedded Systems (HES) use biosensors to capture physiological data, analyse it with advanced algorithms, and provide timely alerts during emergencies. These systems enhance healthcare delivery by supporting diagnosis, early symptom detection, and disease prediction. Despite extensive research on data analysis techniques in healthcare, selecting real-time methods for specific embedded hardware remains challenging. This review aims to summarise and synthesise existing literature to: (a) identify the healthcare challenges addressed by HES and the types of biosignals employed, (b) explore the embedded platforms utilised for implementing HES, and (c) examine the data analysis techniques used for real-time HES applications. A systematic search across three electronic databases (2015-2024), identified 50 relevant studies. These studies span various application domains, biosensing modalities, feature extraction methods, and machine learning and deep learning techniques. Raspberry Pi single-board computers emerged as the most popular embedded platform for implementing AI-based HES. Deep learning, especially convolutional neural networks, dominated, with cardiac health as the primary focus. While the reviewed studies demonstrate promising results, they are often constrained by specific experimental contexts. This review offers a comprehensive overview of real-time data analysis in HES and highlights key opportunities for future research to advance the field.","PeriodicalId":50926,"journal":{"name":"ACM Computing Surveys","volume":"5 1","pages":""},"PeriodicalIF":16.6,"publicationDate":"2026-01-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146048424","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}
Digital twin technology offers a transformative approach to preserve, manage, and enhance tangible cultural heritage through dynamic and immersive digital representations. Despite growing attention, research in this area remains fragmented and lacks systematic synthesis. This study presents a comprehensive systematic literature review analyzing the state of the art in digital twin applications for tangible cultural heritage. A total of 108 studies published between 2002 and August 2025 were synthesized and categorized across three analytical dimensions: user-centric applications, enabling technologies, and maturity levels. The results indicate that most current implementations remain in early maturity stages primarily static digital replicas with limited adaptivity or intelligence. This trend reflects an ongoing transition toward dynamic, interoperable, and data-driven cultural heritage twins. User-centric applications increasingly leverage immersive technologies such as virtual and augmented reality to enhance accessibility and engagement, while enabling technologies like 3D modeling, real-time data integration, and AI-based analytics are still underutilized for intelligent operations. The findings highlight the need for innovation and standardization to advance maturity and scalability. Federated digital twins emerge as a promising pathway for collaborative, secure, and sustainable preservation and access to cultural heritage.
{"title":"Digital Twins for Cultural Heritage: A Systematic Analysis of the State of the Art","authors":"Gizealew Dagnaw, Roberta Capuano, Henry Muccini","doi":"10.1145/3793541","DOIUrl":"https://doi.org/10.1145/3793541","url":null,"abstract":"Digital twin technology offers a transformative approach to preserve, manage, and enhance tangible cultural heritage through dynamic and immersive digital representations. Despite growing attention, research in this area remains fragmented and lacks systematic synthesis. This study presents a comprehensive systematic literature review analyzing the state of the art in digital twin applications for tangible cultural heritage. A total of 108 studies published between 2002 and August 2025 were synthesized and categorized across three analytical dimensions: user-centric applications, enabling technologies, and maturity levels. The results indicate that most current implementations remain in early maturity stages primarily static digital replicas with limited adaptivity or intelligence. This trend reflects an ongoing transition toward dynamic, interoperable, and data-driven cultural heritage twins. User-centric applications increasingly leverage immersive technologies such as virtual and augmented reality to enhance accessibility and engagement, while enabling technologies like 3D modeling, real-time data integration, and AI-based analytics are still underutilized for intelligent operations. The findings highlight the need for innovation and standardization to advance maturity and scalability. Federated digital twins emerge as a promising pathway for collaborative, secure, and sustainable preservation and access to cultural heritage.","PeriodicalId":50926,"journal":{"name":"ACM Computing Surveys","volume":"31 1","pages":""},"PeriodicalIF":16.6,"publicationDate":"2026-01-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146048425","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}
Congestion control (CC) is fundamental for reliable transport layer protocols like TCP. In next-generation networks (NGN), including 5G-Advanced (5GA)/6G, CC algorithms are even more crucial due to the diversity, heterogeneity, and complexity of emerging applications. Achieving performance guarantees while ensuring fairness among NGN applications is increasingly challenging. TCP loss-based congestion control, introduced in the 1980s with packet loss as the primary indicator of “congestion”, has become less effective as the correlation between packet loss and actual congestion has weakened in next-generation networks (NGN). Google developed the Bottleneck Bandwidth and Round-trip propagation time (BBR) algorithm in 2016 as an alternative to loss-based congestion control. This survey reviews the improvement of the BBR algorithm since its first release. We provide a comprehensive algorithmic analysis of BBRv1, BBRv2, and BBRv3 – focusing on performance, fairness, and literature-proposed improvements to address the drawbacks of each BBR-variant. We experimentally evaluate BBRv3 with 5GA use cases, analyzing its ability to utilize bottleneck bandwidth across diverse Quality of Service (QoS) requirements in throughput and latency. Challenges persist in balancing fairness and optimizing buffering capacity for NGN applications. Finally, with the rapid adoption of artificial intelligence (AI) in networks, we discuss BBR enhancements and future intelligent CC.
{"title":"BBR Congestion Control Algorithms: Evolution, Challenges and Future Directions","authors":"Akshita Abrol, Purnima Murali Mohan, Tram Truong-Huu, Mohan Gurusamy","doi":"10.1145/3793537","DOIUrl":"https://doi.org/10.1145/3793537","url":null,"abstract":"Congestion control (CC) is fundamental for reliable transport layer protocols like TCP. In next-generation networks (NGN), including 5G-Advanced (5GA)/6G, CC algorithms are even more crucial due to the diversity, heterogeneity, and complexity of emerging applications. Achieving performance guarantees while ensuring fairness among NGN applications is increasingly challenging. TCP loss-based congestion control, introduced in the 1980s with packet loss as the primary indicator of “congestion”, has become less effective as the correlation between packet loss and actual congestion has weakened in next-generation networks (NGN). Google developed the Bottleneck Bandwidth and Round-trip propagation time (BBR) algorithm in 2016 as an alternative to loss-based congestion control. This survey reviews the improvement of the BBR algorithm since its first release. We provide a comprehensive algorithmic analysis of BBRv1, BBRv2, and BBRv3 – focusing on performance, fairness, and literature-proposed improvements to address the drawbacks of each BBR-variant. We experimentally evaluate BBRv3 with 5GA use cases, analyzing its ability to utilize bottleneck bandwidth across diverse Quality of Service (QoS) requirements in throughput and latency. Challenges persist in balancing fairness and optimizing buffering capacity for NGN applications. Finally, with the rapid adoption of artificial intelligence (AI) in networks, we discuss BBR enhancements and future intelligent CC.","PeriodicalId":50926,"journal":{"name":"ACM Computing Surveys","volume":"7 1","pages":""},"PeriodicalIF":16.6,"publicationDate":"2026-01-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146048427","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}
Anthony Kiggundu, Bin Han, Dennis Krummacker, Hans Schotten
Emerging trends in communication systems, such as network softwarization, functional disaggregation, and multi-access edge computing (MEC), are reshaping both the infrastructural landscape and the application ecosystem. These transformations introduce new challenges for packet transmission, task offloading, and resource allocation under stringent service-level requirements. A key factor in this context is queue impatience, where waiting entities alter their behavior in response to delay. While balking and reneging have been widely studied, this survey focuses on the less explored but operationally significant phenomenon of jockeying, i.e. the switching of jobs or users between queues. Although a substantial body of literature models jockeying behavior, the diversity of approaches raises questions about their practical applicability in dynamic, distributed environments such as 5G and Beyond. This chronicle reviews and classifies these studies with respect to their methodologies, modeling assumptions, and use cases, with particular emphasis on communication systems and MEC scenarios. We argue that forthcoming architectural transformations in next-generation networks will render many existing jockeying models inapplicable. By highlighting emerging paradigms such as MEC, network slicing, and network function virtualization, we identify open challenges, including state dissemination, migration cost, and stability, that undermine classical assumptions. We further outline design principles and research directions, emphasizing hybrid architectures and decentralized decision making as foundations for re-conceptualizing impatience in next-generation communication systems.
{"title":"Chronicles of Jockeying in Queuing Systems","authors":"Anthony Kiggundu, Bin Han, Dennis Krummacker, Hans Schotten","doi":"10.1145/3786318","DOIUrl":"https://doi.org/10.1145/3786318","url":null,"abstract":"Emerging trends in communication systems, such as network softwarization, functional disaggregation, and multi-access edge computing (MEC), are reshaping both the infrastructural landscape and the application ecosystem. These transformations introduce new challenges for packet transmission, task offloading, and resource allocation under stringent service-level requirements. A key factor in this context is queue impatience, where waiting entities alter their behavior in response to delay. While balking and reneging have been widely studied, this survey focuses on the less explored but operationally significant phenomenon of jockeying, i.e. the switching of jobs or users between queues. Although a substantial body of literature models jockeying behavior, the diversity of approaches raises questions about their practical applicability in dynamic, distributed environments such as 5G and Beyond. This chronicle reviews and classifies these studies with respect to their methodologies, modeling assumptions, and use cases, with particular emphasis on communication systems and MEC scenarios. We argue that forthcoming architectural transformations in next-generation networks will render many existing jockeying models inapplicable. By highlighting emerging paradigms such as MEC, network slicing, and network function virtualization, we identify open challenges, including state dissemination, migration cost, and stability, that undermine classical assumptions. We further outline design principles and research directions, emphasizing hybrid architectures and decentralized decision making as foundations for re-conceptualizing impatience in next-generation communication systems.","PeriodicalId":50926,"journal":{"name":"ACM Computing Surveys","volume":"286 1","pages":""},"PeriodicalIF":16.6,"publicationDate":"2026-01-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146042588","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}
Shangheng Du, Jiabao Zhao, Jinxin Shi, Zhentao Xie, Xin Jiang, Yanhong Bai, Liang He
With the rapid development of Large Language Models (LLMs), LLM-based agents have been widely adopted in various fields, becoming essential for autonomous decision-making and interactive tasks. However, current work typically relies on prompt design or fine-tuning strategies applied to vanilla LLMs, which often leads to limited effectiveness in complex agent-related environments. Although numerous recent studies have explored various strategies to optimize LLM-based agents for complex agent tasks, a systematic review summarizing and comparing these methods from a holistic perspective remains lacking. In this survey, we provide a comprehensive review of LLM-based agent optimization approaches, categorizing them into parameter-driven and parameter-free methods. We first focus on parameter-driven optimization, covering fine-tuning-based optimization, reinforcement learning-based optimization, and hybrid strategies, analyzing key aspects such as trajectory data construction, reward function design, and optimization algorithms. Additionally, we briefly discuss parameter-free strategies that optimize agent behavior through prompt engineering and external knowledge retrieval. Finally, we summarize the evaluation for agents, review key applications of LLM-based agents, and discuss the major challenges and promising future directions. A curated collection of the surveyed works is provided at https://github.com/YoungDubbyDu/LLM-Agent-Optimization.
随着大型语言模型（Large Language Models, llm）的快速发展，基于llm的智能体被广泛应用于各个领域，成为自主决策和交互任务的必要条件。然而，目前的工作通常依赖于应用于普通llm的快速设计或微调策略，这通常导致在复杂的代理相关环境中的有效性有限。尽管最近有许多研究探索了各种策略来优化基于llm的复杂代理任务，但从整体角度总结和比较这些方法的系统综述仍然缺乏。在这项调查中，我们提供了基于llm的智能体优化方法的全面回顾，将它们分为参数驱动和无参数方法。我们首先关注参数驱动优化，包括基于微调的优化、基于强化学习的优化和混合策略，分析了轨迹数据构建、奖励函数设计和优化算法等关键方面。此外，我们还简要讨论了通过提示工程和外部知识检索来优化智能体行为的无参数策略。最后，我们总结了对代理的评价，回顾了基于llm的代理的主要应用，并讨论了主要挑战和未来的发展方向。调查作品的精选集在https://github.com/YoungDubbyDu/LLM-Agent-Optimization上提供。
{"title":"A Survey on the Optimization of Large Language Model-based Agents","authors":"Shangheng Du, Jiabao Zhao, Jinxin Shi, Zhentao Xie, Xin Jiang, Yanhong Bai, Liang He","doi":"10.1145/3789261","DOIUrl":"https://doi.org/10.1145/3789261","url":null,"abstract":"With the rapid development of Large Language Models (LLMs), LLM-based agents have been widely adopted in various fields, becoming essential for autonomous decision-making and interactive tasks. However, current work typically relies on prompt design or fine-tuning strategies applied to vanilla LLMs, which often leads to limited effectiveness in complex agent-related environments. Although numerous recent studies have explored various strategies to optimize LLM-based agents for complex agent tasks, a systematic review summarizing and comparing these methods from a holistic perspective remains lacking. In this survey, we provide a comprehensive review of LLM-based agent optimization approaches, categorizing them into parameter-driven and parameter-free methods. We first focus on parameter-driven optimization, covering fine-tuning-based optimization, reinforcement learning-based optimization, and hybrid strategies, analyzing key aspects such as trajectory data construction, reward function design, and optimization algorithms. Additionally, we briefly discuss parameter-free strategies that optimize agent behavior through prompt engineering and external knowledge retrieval. Finally, we summarize the evaluation for agents, review key applications of LLM-based agents, and discuss the major challenges and promising future directions. A curated collection of the surveyed works is provided at https://github.com/YoungDubbyDu/LLM-Agent-Optimization.","PeriodicalId":50926,"journal":{"name":"ACM Computing Surveys","volume":"309 1","pages":""},"PeriodicalIF":16.6,"publicationDate":"2026-01-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146042587","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}
Chunyan Liu, Yan Lei, Huan Xie, Jinping Wang, Yue Yu, David Lo
Learning-based dynamic fault localization techniques play a crucial role in the field of software engineering. These techniques dynamically execute test cases to meticulously extract useful knowledge from the execution information in the program, with the aim of identifying fault locations by leveraging machine learning, deep learning, and large language models. Currently, there is already a flourishing body of research that is intensely focused on learning-based dynamic fault localization. Research literature can be categorized into two main aspects for learning-based dynamic fault localization: data-based enhancements (i.e., the datasets) and model-based enhancements (i.e., the suspiciousness algorithms). Thus, we conduct an extensive literature review on learning-based dynamic fault localization from the aspects of the data task and the model task. Among them, each task is divided into multiple sub-tasks in a systematic manner to comprehensively discuss the details. In addition, we analyze and summarize the datasets and metrics that have been widely used to evaluate the effectiveness of the proposed techniques in recent years, so that researchers can have an intuitive perception of them. We also discuss the present challenges and the directions for future research.
{"title":"Survey on Learning-based Dynamic Fault Localization: From Traditional Machine Learning to Large Language Models","authors":"Chunyan Liu, Yan Lei, Huan Xie, Jinping Wang, Yue Yu, David Lo","doi":"10.1145/3787202","DOIUrl":"https://doi.org/10.1145/3787202","url":null,"abstract":"Learning-based dynamic fault localization techniques play a crucial role in the field of software engineering. These techniques dynamically execute test cases to meticulously extract useful knowledge from the execution information in the program, with the aim of identifying fault locations by leveraging machine learning, deep learning, and large language models. Currently, there is already a flourishing body of research that is intensely focused on learning-based dynamic fault localization. Research literature can be categorized into two main aspects for learning-based dynamic fault localization: data-based enhancements (i.e., the datasets) and model-based enhancements (i.e., the suspiciousness algorithms). Thus, we conduct an extensive literature review on learning-based dynamic fault localization from the aspects of the data task and the model task. Among them, each task is divided into multiple sub-tasks in a systematic manner to comprehensively discuss the details. In addition, we analyze and summarize the datasets and metrics that have been widely used to evaluate the effectiveness of the proposed techniques in recent years, so that researchers can have an intuitive perception of them. We also discuss the present challenges and the directions for future research.","PeriodicalId":50926,"journal":{"name":"ACM Computing Surveys","volume":"55 1","pages":""},"PeriodicalIF":16.6,"publicationDate":"2026-01-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146042586","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}
Benedito Ribeiro Neto, Bianchi Meiguins, Tiago Araújo, Carlos dos Santos
Applications using fiducial markers have evolved across sectors such as industry, health, and education. Markers are effective because their highly distinguishable visual patterns and varied morphologies allow for high-accuracy pose estimation. However, designing a robust fiducial marker system is difficult and requires specific strategies to ensure reliability for applications such as photogrammetry and robot localization. This study aims to address this challenge through a systematic study of 88 articles selected using snowball methodology. This study focused on marker design characteristics to analyze different types of robustness. The goal of this study was to formally define fiducial markers, explore their intrinsic and extrinsic characteristics, and produce a taxonomy covering morphological and algorithmic aspects. The primary outcome is a comprehensive taxonomy and theoretical framework that provides best practices, guiding researchers in developing or employing robust fiducial markers tailored to their specific applications.
{"title":"Artificial Markers: A Comprehensive Systematic Review and Design Framework","authors":"Benedito Ribeiro Neto, Bianchi Meiguins, Tiago Araújo, Carlos dos Santos","doi":"10.1145/3793661","DOIUrl":"https://doi.org/10.1145/3793661","url":null,"abstract":"Applications using fiducial markers have evolved across sectors such as industry, health, and education. Markers are effective because their highly distinguishable visual patterns and varied morphologies allow for high-accuracy pose estimation. However, designing a robust fiducial marker system is difficult and requires specific strategies to ensure reliability for applications such as photogrammetry and robot localization. This study aims to address this challenge through a systematic study of 88 articles selected using snowball methodology. This study focused on marker design characteristics to analyze different types of robustness. The goal of this study was to formally define fiducial markers, explore their intrinsic and extrinsic characteristics, and produce a taxonomy covering morphological and algorithmic aspects. The primary outcome is a comprehensive taxonomy and theoretical framework that provides best practices, guiding researchers in developing or employing robust fiducial markers tailored to their specific applications.","PeriodicalId":50926,"journal":{"name":"ACM Computing Surveys","volume":"40 1","pages":""},"PeriodicalIF":16.6,"publicationDate":"2026-01-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146044844","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}
Hanh Tran, Matej Martinc, Jaya Caporusso, Julien Delaunay, Antoine Doucet, Senja Pollak
Automatic terminology or term extraction (ATE) is a Natural Language Processing (NLP) task intended to automatically identify specialized terms present in domain-specific corpora. As units of knowledge in a specific field of expertise, extracted terms are not only beneficial for several terminographical tasks, but also support and improve several complex downstream tasks, e.g., information retrieval, machine translation, topic detection, and sentiment analysis. ATE systems and datasets annotated for the task at hand have been studied and developed for decades, but more recent approaches have increasingly involved novel neural systems. Despite a large amount of new research on ATE tasks, systematic survey studies covering novel neural approaches are lacking, especially when it comes to the usage of large-scale language models (LLMs). We present a comprehensive survey of neural approaches to ATE, focusing on transformer-based neural models and the recent generative approaches based on LLMs. The study also compares these systems and previous ML-based approaches, which employed feature engineering and non-neural supervised learning algorithms.
{"title":"Recent Advances in Automatic Term Extraction: A Comprehensive Survey","authors":"Hanh Tran, Matej Martinc, Jaya Caporusso, Julien Delaunay, Antoine Doucet, Senja Pollak","doi":"10.1145/3787584","DOIUrl":"https://doi.org/10.1145/3787584","url":null,"abstract":"Automatic terminology or term extraction (ATE) is a Natural Language Processing (NLP) task intended to automatically identify specialized terms present in domain-specific corpora. As units of knowledge in a specific field of expertise, extracted terms are not only beneficial for several terminographical tasks, but also support and improve several complex downstream tasks, e.g., information retrieval, machine translation, topic detection, and sentiment analysis. ATE systems and datasets annotated for the task at hand have been studied and developed for decades, but more recent approaches have increasingly involved novel neural systems. Despite a large amount of new research on ATE tasks, systematic survey studies covering novel neural approaches are lacking, especially when it comes to the usage of large-scale language models (LLMs). We present a comprehensive survey of neural approaches to ATE, focusing on transformer-based neural models and the recent generative approaches based on LLMs. The study also compares these systems and previous ML-based approaches, which employed feature engineering and non-neural supervised learning algorithms.","PeriodicalId":50926,"journal":{"name":"ACM Computing Surveys","volume":"87 1","pages":""},"PeriodicalIF":16.6,"publicationDate":"2026-01-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146044924","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}
Shriyank Somvanshi, Md Monzurul Islam, Amir Rafe, Anannya Ghosh Tusti, Arka Chakraborty, Anika Baitullah, Tausif Islam Chowdhury, Nawaf Alnawmasi, Anandi Dutta, Subasish Das
Mechanistic interpretability seeks to reverse-engineer the internal logic of neural networks by uncovering human-understandable circuits, algorithms, and causal structures that drive model behavior. Unlike post hoc explanations that describe what models do, this paradigm focuses on why and how they compute, tracing information flow through neurons, attention heads, and activation pathways. This survey provides a high-level synthesis of the field-highlighting its motivation, conceptual foundations, and methodological taxonomy rather than enumerating individual techniques. We organize mechanistic interpretability across three abstraction layers- neurons , circuits , and algorithms -and three evaluation perspectives: behavioral , counterfactual , and causal . We further discuss representative approaches and toolchains that enable structural analysis of modern AI systems, outlining how mechanistic interpretability bridges theoretical insights with practical transparency. Despite rapid progress, challenges persist in scaling these analyses to frontier models, resolving polysemantic representations, and establishing standardized causal benchmarks. By connecting historical evolution, current methodologies, and emerging research directions, this survey aims to provide an integrative framework for understanding how mechanistic interpretability can support transparency, reliability, and governance in large-scale AI.
{"title":"Bridging the Black Box: A Survey on Mechanistic Interpretability in AI","authors":"Shriyank Somvanshi, Md Monzurul Islam, Amir Rafe, Anannya Ghosh Tusti, Arka Chakraborty, Anika Baitullah, Tausif Islam Chowdhury, Nawaf Alnawmasi, Anandi Dutta, Subasish Das","doi":"10.1145/3787104","DOIUrl":"https://doi.org/10.1145/3787104","url":null,"abstract":"Mechanistic interpretability seeks to reverse-engineer the internal logic of neural networks by uncovering human-understandable circuits, algorithms, and causal structures that drive model behavior. Unlike post hoc explanations that describe what models do, this paradigm focuses on why and how they compute, tracing information flow through neurons, attention heads, and activation pathways. This survey provides a high-level synthesis of the field-highlighting its motivation, conceptual foundations, and methodological taxonomy rather than enumerating individual techniques. We organize mechanistic interpretability across three abstraction layers- <jats:italic toggle=\"yes\">neurons</jats:italic> , <jats:italic toggle=\"yes\">circuits</jats:italic> , and <jats:italic toggle=\"yes\">algorithms</jats:italic> -and three evaluation perspectives: <jats:italic toggle=\"yes\">behavioral</jats:italic> , <jats:italic toggle=\"yes\">counterfactual</jats:italic> , and <jats:italic toggle=\"yes\">causal</jats:italic> . We further discuss representative approaches and toolchains that enable structural analysis of modern AI systems, outlining how mechanistic interpretability bridges theoretical insights with practical transparency. Despite rapid progress, challenges persist in scaling these analyses to frontier models, resolving polysemantic representations, and establishing standardized causal benchmarks. By connecting historical evolution, current methodologies, and emerging research directions, this survey aims to provide an integrative framework for understanding how mechanistic interpretability can support transparency, reliability, and governance in large-scale AI.","PeriodicalId":50926,"journal":{"name":"ACM Computing Surveys","volume":"1 1","pages":""},"PeriodicalIF":16.6,"publicationDate":"2026-01-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146042589","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}
Massimo Regona, Tan Yigitcanlar, Carol Hon, Melissa Teo
Artificial intelligence (AI) is reshaping industries by enhancing efficiency and accuracy, yet its adoption remains contingent on user trust, which is frequently undermined by concerns over privacy, algorithmic bias, and security vulnerabilities. Trust in AI depends on principles such as transparency, accountability, safety, privacy, robustness, and reliability, all of which are central to user confidence. However, existing studies often overlook the interdependencies among these factors and their collective influence on user engagement. Guided by Trust Theory and a systematic literature review employing the PRISMA protocol, this study examines the trust indicators most relevant to high-stakes applications. The review reveals that transparency and communication are consistently prioritised, while adaptability and affordability remain underexplored, highlighting gaps in current scholarship. Trust in AI evolves as users gain experience with these systems, with reliability, predictability, and ethical alignment emerging as critical determinants. Addressing persistent challenges such as bias, data protection, and fairness is essential for reinforcing trust and enabling broader adoption of AI across industries.
{"title":"Building Trust in Artificial Intelligence: A Systematic Review through the Lens of Trust Theory","authors":"Massimo Regona, Tan Yigitcanlar, Carol Hon, Melissa Teo","doi":"10.1145/3789256","DOIUrl":"https://doi.org/10.1145/3789256","url":null,"abstract":"Artificial intelligence (AI) is reshaping industries by enhancing efficiency and accuracy, yet its adoption remains contingent on user trust, which is frequently undermined by concerns over privacy, algorithmic bias, and security vulnerabilities. Trust in AI depends on principles such as transparency, accountability, safety, privacy, robustness, and reliability, all of which are central to user confidence. However, existing studies often overlook the interdependencies among these factors and their collective influence on user engagement. Guided by Trust Theory and a systematic literature review employing the PRISMA protocol, this study examines the trust indicators most relevant to high-stakes applications. The review reveals that transparency and communication are consistently prioritised, while adaptability and affordability remain underexplored, highlighting gaps in current scholarship. Trust in AI evolves as users gain experience with these systems, with reliability, predictability, and ethical alignment emerging as critical determinants. Addressing persistent challenges such as bias, data protection, and fairness is essential for reinforcing trust and enabling broader adoption of AI across industries.","PeriodicalId":50926,"journal":{"name":"ACM Computing Surveys","volume":"124 1","pages":""},"PeriodicalIF":16.6,"publicationDate":"2026-01-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145986540","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}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：