计算机科学最新文献

{"title":"EMERALD-O: efficient multi-agent reinforcement learning framework for optimised deep learning hyperparameter tuning and selection","authors":"Akhila VH,&nbsp;Anu Mary Chacko,&nbsp;Ponnurangam Kumaraguru","doi":"10.1007/s10489-025-06878-4","DOIUrl":"10.1007/s10489-025-06878-4","url":null,"abstract":"<div><p>Traditional hyperparameter tuning methods, such as Bayesian Optimization and Grid Search, often prove computationally expensive and inefficient for complex deep learning architectures. This paper introduces the Multi-Agent Reinforcement Learning (MARL) framework EMERALD-O to optimize deep learning networks. The MARL-based approach utilizes two specialized agents, Agent1 focuses on data augmentation and Agent 2 on managing the learning rate and optimizer selection. The agents operate within an environment that simulates the model’s training dynamics and uses validation accuracy as the reward signal. Agent performance is enhanced through epsilon-greedy exploration and experience replay mechanisms. EMERALD-O performs favorably 88.59 % with improved classification accuracy and training efficiency. The framework exhibits adaptability to diverse dataset characteristics, underscoring scalability and robustness. The framework was validated on different models built for image classification problem on Efficientnet, VGG16 and VGG19. The results highlight the potential of reinforcement learning to fine-tune complex neural network architectures and suggest that MARL can serve as a powerful tool to improve the performance of deep learning models. EMERALD-O can contribute by advancing the frontier of deep neural optimization, demonstrating that reinforcement learning can fundamentally transform the model-tuning approach. This framework establishes a new paradigm for automated hyperparameter optimization and provides a systematic lens for analyzing the behavior of the deep learning model across various hyperparametric configurations. By quantifying model responsiveness to parameter variations, this approach enables deeper insights into architectural characteristics and performance dynamics, facilitating both the theoretical understanding and practical optimization of deep learning systems.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"55 14","pages":""},"PeriodicalIF":3.5,"publicationDate":"2025-09-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145011748","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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":"Age of Process Information of Mobile Edge Computing Assisted IoT Status Update System Based on Layered Non-Orthogonal Multiple Access and HARQ","authors":"Yue Li,&nbsp;Xiangdong Jia,&nbsp;Hailong Tian,&nbsp;Mangang Xie","doi":"10.1002/ett.70244","DOIUrl":"https://doi.org/10.1002/ett.70244","url":null,"abstract":"&lt;div&gt;\u0000 \u0000 &lt;p&gt;This work focuses on a mobile edge computing (MEC) assisted IoT status update network with multi-objective sensing, which consists of a wireless sensing network, MEC network, and an information receiver (IR). To simultaneously guarantee information freshness and system throughput, a layer-superposed non-orthogonal multiple access (NOMA) HARQ (LS-NOMA-HARQ) scheme is proposed. In the proposed LS-NOMA-HARQ scheme, an entire status update delivery circle consists of multiple rounds for NOMA symbol feedforward transmission. In each round, the source constructs and transmits a NOMA symbol to the AP that first performs feedforward decoding (FD). Each NOMA symbol includes the newly generated packet, termed the primary packet, and the part of the currently failed packet, termed the secondary one. If the received primary packet can not be correctly recovered by the AP, the NOMA symbol is offloaded to the edge server and stored in the buffer of the edge server. The source continuously generates and sends new NOMA symbols to AP until a successful FD occurs. On the contrary, the decoded result is directly delivered to IR, and backtrack decoding (BD) is triggered at the edge server. Then, the edge server successively decodes the previously stored NOMA symbols by using sophisticated successive interference cancellation (SIC), and delivers the recovered packet to IR. Once SIC-based BD fails, it is declared that a circle of LS-NOMA-HARQ status update delivery completes. Because the primary and secondary packets are independently modulated and superposed in the MAC layer, the proposed LS-NOMA-HARQ outperforms the layer-coded HARQ scheme that is executed in the physical layer. Moreover, this work also considers the two cases of finite and infinite buffer size at the edge server, and truncated HARQ is used for the retransmission of a single NOMA symbol. Under the finite buffer size case, the circle-shift preemption is used at the buffer edge server. The edge service follows exponentially distributed processes due to the user schedule and can be interrupted by one In-Out process due to the energy computation at the edge server, which results in a huge data processing delay at the edge server. To characterize this specific issue, this work investigates the age of process information (AoPI). In addition, considering the joint impact of both FD and BD, two modified AoPI metrics, that is, FD-based AoPI (FD-AoPI) and BD based AoPI (BD-AoPI), are proposed. The FD-AoPI is defined as the elapsed time since the generation of the last successfully feedforward decoded update, but the BD-AoPI is based on the classical definition of AoI. While the FD-AoPI simultaneously captures both throughput and information freshness, the BD-AoPI results in a loss in information timeliness. With the statistical characterization of related statistics, the closed-form expressions of both FD-AoPI and BD-AoPI are derived. The simulated and numerical results give insigh","PeriodicalId":23282,"journal":{"name":"Transactions on Emerging Telecommunications Technologies","volume":"36 9","pages":""},"PeriodicalIF":2.5,"publicationDate":"2025-09-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145012506","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"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":"Introduction to the special issue on innovative applications of artificial intelligence (IAAI 2025)","authors":"Serdar Kadıoğlu,&nbsp;Sean McGregor,&nbsp;Jan Seyler","doi":"10.1002/aaai.70027","DOIUrl":"https://doi.org/10.1002/aaai.70027","url":null,"abstract":"<p>This year's <i>innovative applications of AI</i> special issue features AI systems deployed in real-world settings, from enterprise platforms to public services, demonstrating both technical rigor and measurable benefits for organizations and society. The eight selected articles span enterprise reliability, cybersecurity, aerospace, education, healthcare logistics, government services, and scalable AI strategy. Collectively, these works illustrate how AI is progressing from research prototypes to systems that organizations now rely on for critical decisions, offering lessons learned for both researchers and practitioners.</p>","PeriodicalId":7854,"journal":{"name":"Ai Magazine","volume":"46 3","pages":""},"PeriodicalIF":3.2,"publicationDate":"2025-09-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145012548","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"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":"Multimodal Medical Image Fusion With UNet-Based Multi-Scale Transformer Networks","authors":"Qingwu Fu,&nbsp;Jianxuan Zhou,&nbsp;Jiao Du,&nbsp;Kai Lin,&nbsp;Bowen Zhong,&nbsp;Haoran Tang,&nbsp;Yiting Chen","doi":"10.1002/ima.70193","DOIUrl":"https://doi.org/10.1002/ima.70193","url":null,"abstract":"<div>\u0000 \u0000 <p>Multimodal medical image fusion can generate medical images that contain both functional metabolic information and structural tissue details, thereby providing doctors with more comprehensive information. Current deep learning-based methods often employ convolutional neural networks (CNNs) for feature extraction. However, CNNs exhibit limitations in capturing global contextual information compared to Transformers. Moreover, single-scale networks fail to exploit the complementary information between different scales, which limits their ability to fully capture rich image features and results in suboptimal fusion performance. To address these limitations, this paper proposes a multimodal medical image fusion method with UNet-based multi-scale Transformer network. First, we design a UNet-based encoder that incorporates a lightweight Transformer model, PVTv2, to extract multi-scale features from both MRI and SPECT images. To enhance the structural details of MRI images, we introduce the Edge-Guided Attention Module. Additionally, we propose an objective function that combines structural and pixel-level losses to optimize the proposed network. We perform both qualitative and quantitative experiments on mainstream datasets, and the results demonstrate that the proposed method outperforms several representative methods. In addition, we extend the proposed method to other biomedical functional and structural image fusion tasks, and the results show that the proposed method has good generalization capability.</p>\u0000 </div>","PeriodicalId":14027,"journal":{"name":"International Journal of Imaging Systems and Technology","volume":"35 5","pages":""},"PeriodicalIF":2.5,"publicationDate":"2025-09-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145012628","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"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":"On the design of linear time varying model predictive control for trajectory stabilization","authors":"Nicolas Kessler ,&nbsp;Lorenzo Fagiano","doi":"10.1016/j.automatica.2025.112577","DOIUrl":"10.1016/j.automatica.2025.112577","url":null,"abstract":"<div><div>Stabilizing a reference trajectory of a nonlinear system is a recurrent, non-trivial task in control engineering. A common approach is to linearize the dynamics along the trajectory, thus deriving a linear-time-varying (LTV) model, and to design a model predictive controller (MPC), which results to be computationally efficient, since only convex programs need to be solved in real time, while retaining constraint handling capabilities. Building on recent developments in gain-scheduling control design, where linearization errors and tracking error bounds are considered, a new approach to derive such LTV-MPC controllers is presented. The method addresses the systematic derivation of a suitable terminal cost. The resulting MPC law is tube-based, exploiting the co-designed auxiliary gain-scheduled controller. Computational and implementation aspects are discussed as well, and the resulting hierarchical method is demonstrated both in simulation and in experiments with a small drone with fast dynamics and limited embedded computational capacity.</div></div>","PeriodicalId":55413,"journal":{"name":"Automatica","volume":"183 ","pages":"Article 112577"},"PeriodicalIF":5.9,"publicationDate":"2025-09-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145010785","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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 compute-efficient Byzantine-robust federated learning with fully homomorphic encryption","authors":"Siyang Jiang, Hao Yang, Qipeng Xie, Chuan Ma, Sen Wang, Zhe Liu, Tao Xiang, Guoliang Xing","doi":"10.1038/s42256-025-01107-6","DOIUrl":"https://doi.org/10.1038/s42256-025-01107-6","url":null,"abstract":"<p>In highly regulated domains such as finance and healthcare, where stringent data-sharing constraints pose substantial obstacles, federated learning (FL) has emerged as a transformative paradigm in distributed machine learning, facilitating collaborative model training, preserving data decentralization and upholding governance standards. Despite its advantages, FL is vulnerable to poisoning attacks during central model aggregation, prompting the development of Byzantine-robust FL systems that use robust aggregation rules to counter malicious attacks. However, neural network models in such systems are susceptible to unintentionally memorizing and revealing individual training instances, thereby introducing substantial information leakage risks, as adversaries may exploit this vulnerability to reconstruct sensitive data through model outputs transmitted over the air. Existing solutions fall short of providing a viable Byzantine-robust FL system that is completely secure against information leakage and is computationally efficient. To address these concerns, we propose Lancelot, an efficient and effective Byzantine-robust FL framework that uses fully homomorphic encryption to safeguard against malicious client activities. Lancelot introduces a mask-based encrypted sorting mechanism that overcomes the limitations of multiplication depth in ciphertext sorting with zero information leakage. It incorporates cryptographic enhancements like lazy relinearization, dynamic hoisting and GPU acceleration to ensure practical computational efficiency. Extensive experiments demonstrate that Lancelot surpasses existing approaches, achieving a 20-fold enhancement in processing speed.</p>","PeriodicalId":48533,"journal":{"name":"Nature Machine Intelligence","volume":"16 1","pages":""},"PeriodicalIF":23.8,"publicationDate":"2025-09-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145009023","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":"Reliable Saturation Control for Multiple Asynchronous Switched Positive Systems With Adaptive Event-Triggered Control","authors":"Hongyuan Ma,&nbsp;Le Zhang,&nbsp;Hong Yang,&nbsp;Ying Zhao","doi":"10.1049/cth2.70059","DOIUrl":"https://doi.org/10.1049/cth2.70059","url":null,"abstract":"<p>This paper investigates the L1 gain stability problem of reliable control for positive systems with input saturation under multi-asynchronous switching. Firstly, by constructing a system state observer and integrating it with an output feedback control strategy, the input variables for the system controller were obtained, and a reliable controller with input saturation was designed. Secondly, to prevent data accumulation, an adaptive event-triggered control strategy that ensures the non-negativity requirements of positive systems is introduced between the observer and the system state. This strategy can adjust the tightness of the event-triggering process, which not only improves control efficiency but also reduces the risk of the Zeno effect. The following describes a switching strategy based on event-triggered control. Under the guidance of a time-varying mode-dependent average dwell-time switching strategy, the multi-asynchronous delay problem of sub-observers and sub-controllers with respect to subsystems is addressed, leading to a closed-loop control system based on error feedback. By constructing co-positive Lyapunov function, sufficient conditions for the positivity of the system under both synchronous- and asynchronous-switching are provided, and the L1 gain stability of the system in both synchronous and asynchronous intervals is verified. Finally, the significance of the proposed method is validated through an example.</p>","PeriodicalId":50382,"journal":{"name":"IET Control Theory and Applications","volume":"19 1","pages":""},"PeriodicalIF":2.3,"publicationDate":"2025-09-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ietresearch.onlinelibrary.wiley.com/doi/epdf/10.1049/cth2.70059","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145012137","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Meeting companies’ innovative requirements on online technology trading platforms: A novel large language model-based framework","authors":"Qingyu Xu,&nbsp;Zhaobin Liu,&nbsp;Jian Ma","doi":"10.1016/j.ipm.2025.104392","DOIUrl":"10.1016/j.ipm.2025.104392","url":null,"abstract":"<div><div>Online technology trading platforms (OTTPs) are critical for companies to publish technology requirements and identify solutions like patents. However, semantic gaps persist between market-driven needs and technical supply texts, which traditional methods fail to bridge. While large language models (LLMs) show promise, their effectiveness in OTTPs is limited by hallucination and temporal unawareness. We propose an LLM framework integrating the Hypothetical Document Embedding (HyDE) framework, where we generate pseudo-supply texts based on technical requirements. These texts are then matched with candidate patents using similarity calculations. To reduce hallucination, we use industry-specific knowledge graphs to guide the text generation process and introduce a self-reflective mechanism to refine the generated texts. To address the lack of time awareness, we enhance the knowledge graph with timestamps, turning it into a temporal knowledge graph. Additionally, we introduce the TPPR (Temporal Personalized PageRank) algorithm to improve the relevance of generated texts. Experiments show that our framework performs better than existing methods in Recall, Precision, and Mean Reciprocal Rank (MRR). This framework advances technology forecasting by enabling dynamic patent matching, offering organizations actionable insights for R&amp;D investments. ​By reducing mismatches and innovation cycle times, it supports sustainable technology transfer—highlighting implications for AI governance in evolving innovation ecosystems.</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"63 2","pages":"Article 104392"},"PeriodicalIF":6.9,"publicationDate":"2025-09-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145010927","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"管理学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A distributed influence measurement algorithm in leader–follower networks","authors":"Hyung-Gon Lee ,&nbsp;Jeong-Min Ma ,&nbsp;Nam-Jin Park ,&nbsp;Hyo-Sung Ahn","doi":"10.1016/j.sysconle.2025.106230","DOIUrl":"10.1016/j.sysconle.2025.106230","url":null,"abstract":"<div><div>This study proposes a <em>vector-wise step-sized consensus dynamics (VSCD)</em> for distributed networks represented by positively weighted leader–follower graphs. Unlike traditional discrete consensus dynamics, VSCD employs node-specific vector step sizes, enabling faster convergence. We define an influence matrix in continuous consensus dynamics and extend it to a discrete influence matrix in VSCD, demonstrating equivalent convergence properties under specific vector step size conditions. To facilitate the application of VSCD in distributed networks, we analyze the maximum boundary vector step size conditions using graph-theoretic methods. Building on this formulation, we propose a fully <em>distributed influence measurement algorithm (DIMA)</em>, which enables each node in a distributed network to determine its valid influence nodes and their corresponding influence using only local information, without requiring global parameters. The effectiveness and scalability of the proposed methods are validated through simulations.</div></div>","PeriodicalId":49450,"journal":{"name":"Systems & Control Letters","volume":"205 ","pages":"Article 106230"},"PeriodicalIF":2.5,"publicationDate":"2025-09-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145010955","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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":"Adaptive landscape-aware repelling restart covariance matrix adaptation-evolution strategy for multimodal and global optimization","authors":"Xikang Wang,&nbsp;Tongxi Wang,&nbsp;Hua Xiang","doi":"10.1016/j.swevo.2025.102143","DOIUrl":"10.1016/j.swevo.2025.102143","url":null,"abstract":"<div><div>In multimodal optimization using Covariance Matrix Adaptation-Evolution Strategy (CMA-ES), redundant restarts are caused by repeated convergence to previously explored local basins, which leads to significant computational resource waste. To address this problem, previous research proposed the concept of Repelling Restart and developed RR-CMA-ES, but issues remain regarding rigid repulsion and gradient information of local basin structures. Building on this foundation, we propose an Adaptive Landscape-aware Repelling Restart CMA-ES (ALR-CMA-ES) that enhances the original RR-CMA-ES through three key improvements: 1) A fitness sensitive dynamic exclusion mechanism that adaptively adjusts tabu region radius based on local optimality and convergence frequency, prioritizing avoidance of high-quality basins; 2) A covariance matrix mechanism preserving convergence history to geometrically align hyper-ellipsoidal exclusion regions with explored local basin landscapes; 3) A Boltzmann-like probabilistic acceptance scheme incorporating exclusion regions, permit- ting controlled exploration near tabu boundaries. Experiments on the BBOB benchmark demonstrate that ALR-CMA-ES outperforms RR-CMA-ES in 90% of tested problems spanning 2D to 50D. This method provides a practical solution for expensive black-box optimization by systematically integrating landscape topology awareness into tabu mechanisms, while proposing a new solution for multimodal optimization problems.</div></div>","PeriodicalId":48682,"journal":{"name":"Swarm and Evolutionary Computation","volume":"99 ","pages":"Article 102143"},"PeriodicalIF":8.5,"publicationDate":"2025-09-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145011296","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}