IEEE Transactions on Emerging Topics in Computing最新文献

{"title":"DeepNcode: Encoding-Based Protection Against Bit-Flip Attacks on Neural Networks","authors":"Patrik Velčický;Jakub Breier;Mladen Kovačević;Xiaolu Hou","doi":"10.1109/TETC.2026.3704626","DOIUrl":"https://doi.org/10.1109/TETC.2026.3704626","url":null,"abstract":"Fault injection attacks are a potent threat against embedded implementations of neural network models. Several attack vectors have been proposed, such as misclassification, model extraction, and trojan/backdoor planting. Most of these attacks work by flipping bits in the memory where quantized model parameters are stored. In this paper, we introduce an encoding-based protection method against bit-flip attacks on neural networks, titled DeepNcode. We experimentally evaluate our proposal with several publicly available models and datasets, by using state-of-the-art bit-flip attacks: BFA, T-BFA, and TA-LBF. Our results show an increase in protection margin of up to <inline-formula><tex-math>$7.6times$</tex-math></inline-formula> for <inline-formula><tex-math>$4-$</tex-math></inline-formula>bit and <inline-formula><tex-math>$12.4times$</tex-math></inline-formula> for <inline-formula><tex-math>$8-$</tex-math></inline-formula>bit quantized networks. Memory overheads start at 50% of the original network size, while the time overheads are negligible. Moreover, DeepNcode does not require retraining and does not change the original accuracy of the model.","PeriodicalId":13156,"journal":{"name":"IEEE Transactions on Emerging Topics in Computing","volume":"14 3","pages":"1181-1196"},"PeriodicalIF":4.8,"publicationDate":"2026-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11575728","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148902817","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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":"UPForest: Enabling Random Forest Training on Real Processing-in-Memory Systems","authors":"Yi-Peng Hsiao;Chun-Feng Wu","doi":"10.1109/TETC.2026.3709681","DOIUrl":"https://doi.org/10.1109/TETC.2026.3709681","url":null,"abstract":"Random Forest (RF) is a popular ensemble learning method widely used for classification and regression tasks in machine learning. However, training RF models is time-consuming in traditional processor-centric systems due to extensive data movement. To address this, we propose training RF with UPMEM, a Processing-in-Memory (PIM) architecture that places processing units closer to memory. This architecture aims to reduce data transfer costs and leverages extensive parallelism to improve computational efficiency. However, UPMEM faces several constraints, including limited isolated memory capacity and restricted computing power per unit. To overcome these limitations, we present UPForest, a co-design solution that enables RF training on UPMEM while effectively addressing these constraints. Evaluation results show that UPForest achieves significant improvements in execution time and reduces data movement compared to the popular RF implementation, Ranger.","PeriodicalId":13156,"journal":{"name":"IEEE Transactions on Emerging Topics in Computing","volume":"14 3","pages":"958-968"},"PeriodicalIF":4.8,"publicationDate":"2026-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148902946","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":"Hyperspectral Imaging-Based Grain Quality Assessment With Limited Labelled Data","authors":"Priyabrata Karmakar;Manzur Murshed;Shyh Wei Teng","doi":"10.1109/TETC.2026.3703954","DOIUrl":"https://doi.org/10.1109/TETC.2026.3703954","url":null,"abstract":"Hyperspectral imaging (HSI) enables non-invasive grain quality assessment, but the large labelled datasets required for deep convolutional neural network (DCNN) training are costly and time-consuming to obtain. In this paper, we address this challenge using few-shot learning (FSL) with two key innovations: (1) collective class prototypes (CCPs) for robust inference and (2) a modified squeeze-excitation attention mechanism, combining adaptive average and max pooling, for improved spectral feature representation. Our prototypical network classifier achieved 97.75% accuracy for classifying eight grain types with only 360 labelled samples per class (2,880 total), demonstrating 3.66× faster inference than support-set-based approaches. Critically, when evaluated on novel grain varieties unseen during training, the model achieved 83.89% accuracy with 5-shot learning. These results demonstrate FSL as a practical solution for rapid, accurate grain quality assessment, where labelled data is scarce and new grain grades emerge frequently.","PeriodicalId":13156,"journal":{"name":"IEEE Transactions on Emerging Topics in Computing","volume":"14 3","pages":"1241-1252"},"PeriodicalIF":4.8,"publicationDate":"2026-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148903175","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":"IEEE Transactions on Emerging Topics in Computing Publication Information","authors":"","doi":"10.1109/TETC.2026.3728706","DOIUrl":"https://doi.org/10.1109/TETC.2026.3728706","url":null,"abstract":"","PeriodicalId":13156,"journal":{"name":"IEEE Transactions on Emerging Topics in Computing","volume":"14 3","pages":"C2-C2"},"PeriodicalIF":4.8,"publicationDate":"2026-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11679566","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148903199","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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":"Vector Field-Guided Circumnavigation for an Unmanned Surface Vehicle With Boundary and Obstacle Constraints","authors":"Nailong Wu;Jigang Wang;Yueying Wang;Shuping He","doi":"10.1109/TETC.2025.3547104","DOIUrl":"https://doi.org/10.1109/TETC.2025.3547104","url":null,"abstract":"This article aims to tackle the challenge of rapid target circumnavigation by unmanned surface vehicles (USVs) under boundary and obstacle constraints through leveraging an emerging computing architecture specifically tailored for unmanned systems. Owing to the underactuation of USV dynamic systems and the presence of various time-varying surface disturbances, attaining rapid and accurate position control along with maneuverability poses a significant challenge to USVs. Therefore, when an USV maneuvers around a target in a confined search area with obstacles, it is essential for the guidance law to consider collision risks and environmental disturbances. This article proposes a modified Lyapunov vector field with heading deviation compensation to circle the target in complex search regions. The vector field is designed to increase the radial velocity of the USV and accelerate convergence towards the target circle during the pursuit phase. By combining the proposed Lyapunov vector field with a boundary vector field, the USV can swiftly avoid boundaries and dynamic obstacles and then continue circling the target. The proposed computing architecture extends the application scope of the target circumnavigation to practical water surface environments, which is validated by extensive simulations and rigorous experiments.","PeriodicalId":13156,"journal":{"name":"IEEE Transactions on Emerging Topics in Computing","volume":"14 3","pages":"907-918"},"PeriodicalIF":4.8,"publicationDate":"2026-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148902683","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":"RepQuant: Toward Accurate Post-Training Quantization of Large Transformer Models via Scale Reparameterization","authors":"Zhikai Li;Xuewen Liu;Jing Zhang;Qingyi Gu","doi":"10.1109/TETC.2026.3704498","DOIUrl":"https://doi.org/10.1109/TETC.2026.3704498","url":null,"abstract":"Large transformer models have demonstrated remarkable success across vision, language, and multi-modal domains. However, due to their complex structures, they are criticized for high latency and resource consumption during inference. Post-training quantization (PTQ), which requires only a small dataset for calibration and avoids end-to-end retraining, is a promising solution for compressing these large models. Regrettably, existing PTQ methods typically exhibit non-trivial performance loss, especially in low-bit settings. We find that the performance bottleneck stems from over-consideration of hardware compatibility in the quantization process, compelling them to reluctantly employ simple quantizers, albeit at the expense of accuracy. With the above insights, we propose RepQuant, a novel PTQ framework with quantization-inference decoupling paradigm to address the above issues, and it is compatible with multiple transformer variants. RepQuant employs complex quantizers in the quantization process and simplified quantizers in the inference process, and performs mathematically equivalent transformations between the two through quantization scale reparameterization, thus ensuring both accurate quantization and efficient inference. More specifically, we focus on two components with extreme distributions: LayerNorm activations with severe inter-channel variations and Softmax activations with power-law characteristics. Initially, we apply channel-wise quantization and log<inline-formula><tex-math>$sqrt{2}$</tex-math></inline-formula> quantization, respectively, which are tailored to their distributions. In particular, for the former, we introduce a learnable per-channel dual clipping scheme, which is designed to efficiently identify outliers in the unbalanced activations with fine granularity. Then, we reparameterize the scales to hardware-friendly layer-wise quantization and log2 quantization for inference. Moreover, quantized weight reconstruction is seamlessly integrated into the above procedure to further push the performance limits. Extensive experiments are performed on different large-scale transformer variants on multiple tasks, including vision, language, and multi-modal transformers, and RepQuant encouragingly demonstrates significant performance advantages and strong robustness.","PeriodicalId":13156,"journal":{"name":"IEEE Transactions on Emerging Topics in Computing","volume":"14 3","pages":"1046-1061"},"PeriodicalIF":4.8,"publicationDate":"2026-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148902883","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":"d-TV-DBN: A Hierarchical and Distributed Architecture for Scalable Cyber-Physical Anomaly Detection With Bayesian Networks","authors":"Simone Guarino;Francesco Vitale;Ernesto Del Prete;Luca Faramondi;Nicola Mazzocca;Roberto Setola","doi":"10.1109/TETC.2025.3573051","DOIUrl":"https://doi.org/10.1109/TETC.2025.3573051","url":null,"abstract":"The wide availability of diverse data from monitoring industrial cyber-physical systems enables data-driven anomaly detection. In particular, the scientific literature has addressed explainable cyber-physical anomaly detection using Bayesian networks, involving decision fusion from an ensemble of detectors. However, the aggregation of multiple decisions from different detectors within a single Bayesian network may negatively affect anomaly detection performance due to the heterogeneity of distributed environments. Furthermore, such decision fusion may be affected by scalability issues due to Bayesian inference being NP-hard. To address these issues, we present <italic>d</i>-TV-DBN, a hierarchical and distributed architecture to scale cyber-physical anomaly detection with the state-of-the-art ensemble decision fusion technique based on Time-Varying Dynamic Bayesian networks. We provide a Proof-of-Concept evaluation of <italic>d</i>-TV-DBN and evaluate its scalability and detection performance, comparing the results with several ensemble decision fusion approaches. The results demonstrate that the proposed <italic>d</i>-TV-DBN outperforms the current state-of-the-art decision fusion techniques while maintaining a remarkably low mean inference time for reaching the final decision.","PeriodicalId":13156,"journal":{"name":"IEEE Transactions on Emerging Topics in Computing","volume":"14 3","pages":"867-878"},"PeriodicalIF":4.8,"publicationDate":"2026-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11018853","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148902685","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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":"LiCoCAD: A Lightweight Model–LLM Collaborative Framework for CAD Reverse Engineering From Point Clouds","authors":"Xu Xu;Jiaxing Lu;Yan Shi;Fangwei Ning","doi":"10.1109/TETC.2026.3707250","DOIUrl":"https://doi.org/10.1109/TETC.2026.3707250","url":null,"abstract":"Point-cloud-based CAD reverse engineering aims to recover editable parametric models from scanned geometry. Existing methods remain limited by inaccurate effective-parameter prediction, redundant point-cloud inputs, and costly task-specific training. To address these limitations, we propose LiCoCAD, a lightweight model–LLM collaborative framework that reconstructs point clouds as CADQuery code. It uses lightweight modules to generate key constraints and prompts for LLM-based CAD reconstruction. The lightweight CAD-Coder predicts command sequences as structural constraints, while a training-free geometry keypoint extractor converts scanned point clouds into compact coordinate prompts as geometric constraints. These structural and geometric constraints guide commercial LLMs to reason about CAD parameters and generate CADQuery code. This design avoids task-specific LLM fine-tuning and trains only the lightweight CAD-Coder, thereby shifting the main training burden to lightweight task-specific model training. During inference, a CD-feedback refinement stage enforces code validity and refines geometric parameters. Experiments on three datasets show that LiCoCAD achieves higher reconstruction accuracy and robustness than existing methods. Real-world industrial cases further show that the generated CADQuery code remains editable. These results suggest that the lightweight model–LLM collaborative paradigm provides a scalable solution for CAD reverse engineering and is transferable to broader engineering tasks.","PeriodicalId":13156,"journal":{"name":"IEEE Transactions on Emerging Topics in Computing","volume":"14 3","pages":"1062-1074"},"PeriodicalIF":4.8,"publicationDate":"2026-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148902711","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":"Resource-Aware Federated Generative AI With Adaptive Model Training and Intelligent Data Selection for Efficient Disease Diagnosis","authors":"Sakshi Patni;Sungpil Woo;Joohyung Lee","doi":"10.1109/TETC.2026.3696330","DOIUrl":"https://doi.org/10.1109/TETC.2026.3696330","url":null,"abstract":"Healthcare federated learning faces institutional diversity, data constraints, and computing resource discrepancies among other difficulties. These problems call for creative solutions for fair and effective distributed medical image analysis that improves model performance while handling the unique data features in different healthcare environments. This paper presents RAGE-FED (Resource Aware GEneration for FedErated Disease Diagnosis), a new paradigm meant to address these important problems. Emphasising computational efficiency, communication overhead reduction, and energy saving, RAGE-FED seeks to maximize performance across healthcare facilities while ensuring diagnostic accuracy and privacy. To achieve these objectives the framework makes use of advanced data generation methods and adaptive training systems. To effectively use the different processing capacity of the participating nodes, it uses dynamic model complexity adjustment and adaptive learning rate scheduling. To protect private patient data, RAGE-FED additionally uses privacy-preserving techniques like differential privacy and safe aggregation. By use of synthetic data creation, the framework also solves data shortage and imbalance problems. The performance evaluation on the Chest X-Ray Pneumonia dataset confirms the effectiveness of RAGE-FED in diverse healthcare settings, showing gains of over 21.47% in efficiency, communication, and energy use compared to state-of-the-art methods. Extended experiments on additional datasets further validate generalizability, with improvements of 20.13% on Brain Tumor MRI, and 17.85% on HAM10000 skin lesion images. Together with ablation and sensitivity analyses, these results demonstrate the robustness, scalability, and broad applicability of the proposed framework.","PeriodicalId":13156,"journal":{"name":"IEEE Transactions on Emerging Topics in Computing","volume":"14 3","pages":"1135-1152"},"PeriodicalIF":4.8,"publicationDate":"2026-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148902733","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":"Enhancing Copyright Protection in Federated Learning: A Hierarchical Watermarking Approach","authors":"Wei Hou;Yao Ge;Wenqiong Zhang;Yun Li","doi":"10.1109/TETC.2026.3708428","DOIUrl":"https://doi.org/10.1109/TETC.2026.3708428","url":null,"abstract":"Copyright protection in Federated Learning (FL) environments faces some challenges, such as fidelity, validity, robustness, and persistence, due to the large number of diverse FL clients. To address these issues, we propose a Grouped Clients Hierarchical Watermarking (GCHW) scheme, which consists of client grouping, watermark generation, watermark embedding and watermark verification. The client grouping is to group clients based on shared characteristics. In watermark generation, a hierarchical framework is proposed to establish dependencies between the group and client watermarks by embedding part of the group watermark information into each client watermark. Moreover, our GCHW scheme innovatively embeds binary watermarks into the frequency domain of model parameters at both the group and client levels during the watermark embedding phase. Experimental results demonstrate that GCHW effectively maintains fidelity and the validity of the watermarks is validated across independent and identically distributed (IID) and non-IID scenarios. GCHW also exhibits strong robustness against common attacks and defensive aggregation algorithms. In addition, GCHW shows good persistence in dynamic FL environments.","PeriodicalId":13156,"journal":{"name":"IEEE Transactions on Emerging Topics in Computing","volume":"14 3","pages":"943-957"},"PeriodicalIF":4.8,"publicationDate":"2026-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148902815","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}