检索结果-内蒙古大学图书馆

Towards Lifelong Learning of Large Language Models: A Survey

ACM Computing Surveys 2025年第8期57卷 1-35页

作者： Zheng, Junhao Qiu, Shengjie Shi, Chengming Ma, Qianli School of Computer Science and Engineering South China University of Technology Guangzhou China

As the applications of large language models (LLMs) expand across diverse fields, their ability to adapt to ongoing changes in data, tasks, and user preferences becomes crucial. Traditional training methods with static datasets are inadequate for coping with the dynamic nature of real-world information. Lifelong learning, or continual learning, addresses this by enabling LLMs to learn continuously and adapt over their operational lifetime, integrating new knowledge while retaining previously learned information and preventing catastrophic forgetting. Our survey explores the landscape of lifelong learning, categorizing strategies into two groups based on how new knowledge is integrated: Internal Knowledge, where LLMs absorb new knowledge into their parameters through full or partial training, and External Knowledge, which incorporates new knowledge as external resources such as Wikipedia or APIs without updating model parameters. The key contributions of our survey include: (1) introducing a novel taxonomy to categorize the extensive literature of lifelong learning into 12 scenarios;(2) identifying common techniques across all lifelong learning scenarios and classifying existing literature into various technique groups;(3) highlighting emerging techniques such as model expansion and data selection, which were less explored in the pre-LLM era. © 2025 Copyright held by the owner/author(s). Publication rights licensed to ACM.

关键词： Federated learning

来源：评论

学校读者我要写书评

暂无评论

Semi-Supervised Skin Lesion Segmentation Based on Pseudo-Labels

IAENG International Journal of Computer Science

引用

IAENG International Journal of computer science 2025年第2期52卷 325-332页

作者： Mu, Bo Wei, JingXin Zhang, Yujun School of Computer Science and Software Engineering University of Science and Technology Liaoning Anshan114051 China

In recent years, deep learning has significantly advanced skin lesion segmentation. However, annotating medical image data is specialized and costly, while obtaining unlabeled medical data is easier. To address this challenge, we propose a semi-supervised learning strategy to improve segmentation accuracy by combining a small amount of annotated data with a larger volume of unlabeled data. Our approach employs a teacher-student model framework. In this framework, the teacher model generates pseudo-labels for the unlabeled data, and the student model is trained using both these pseudo-labels and the limited true labels. To improve the student model’s learning capacity, we introduce auxiliary segmentation heads that provide joint guidance during training. We use the crossentropy (CE) loss function to quantify the discrepancies between the segmentation outputs of the main head and auxiliary heads. Since pseudo-labels generated by the teacher model may contain noise, we developed a mechanism to identify and exclude uncertain regions in each unlabeled image. This reduces pseudolabel noise and mitigates its negative impact on the student model. Our method demonstrates significant improvements in skin lesion segmentation on the publicly available ISIC2018 dataset, achieving Dice coefficients of 87.84% and 88.73% with only 5% and 10% of the total annotated data, respectively, outperforming existing methods. © (2025), (International Association of Engineers). All rights reserved.

关键词： Self-supervised learning

来源：评论

学校读者我要写书评

暂无评论

Vision-Text Bidirectional Collaborative Image Captioning Algorithm

IAENG International Journal of Computer Science

引用

IAENG International Journal of computer science 2025年第2期52卷 515-523页

作者： Li, Mei-Qi Zhou, Zi-Wei School of Computer Science and Software Engineering University of Science and Technology LiaoNing Anshan114051 China

Image captioning is an interdisciplinary research hotspot at the intersection of computer vision and natural language processing, representing a multimodal task that integrates core technologies from both fields. This task requires the use of computer vision techniques to analyze and extract key visual features from images, followed by the application of natural language processing techniques to generate descriptive text that is syntactically and semantically aligned with human cognition. This process poses a significant challenge for computers. Existing models mostly ignore the relative positional information of visual objects and struggle to efficiently capture the complex relationships between visual and textual data. To address these challenges, we propose a vision-to-text bidirectional collaborative image captioning method. This approach extracts both visual features and positional information of objects, allowing the model to better understand the spatial relationships between objects. The CEW word embedding approach encodes textual information more profoundly, enhancing semantic expression and contextual understanding. In the decoding phase, a bidirectional cross-attention mechanism strengthens the interaction between vision and text, leading to improved accuracy in image understanding. The model is trained and tested on the MSCOCO 2014 dataset and compared with several popular models. Experimental results demonstrate that the proposed method achieves significant improvements on the CIDEr and BLEU-1 evaluation metrics with an increase of 1.5 and 1.1, respectively. In addition, we conduct ablation experiments, quantitative analysis, and qualitative analysis to comprehensively validate the effectiveness and stability of the proposed algorithm. © (2025), (International Association of Engineers). All rights reserved.

关键词： Embeddings

来源：评论

学校读者我要写书评

暂无评论

Cryptography with optimal deep learning-based authentication scheme for preserving anonymity in telecare medical information system

引用

Multimedia Tools and Applications 2025年第10期84卷 8311-8330页

作者： Sheik, Syed Amma Durai, Saleem School of Computer Science and Engineering Vellore Institute of Technology Tamil Nadu India

The Telecare Medicine Information System (TMIS) revolutionizes healthcare delivery by integrating medical equipment and sensors, facilitating proactive and cost-effective services. Accessible online, TMIS empowers patients and practitioners to manage medical data efficiently, ensuring timely care provision and streamlined record maintenance. Recognizing security challenges inherent in this interconnected healthcare landscape, researchers have devised a two-factor authentication solution, increasingly leveraging password-based authentication via smart cards. This article presents the development of a tailored remote authentication system, the Deep Learning-Based Online Threat Detection with Key Agreement (DLOTDKA-MAP), designed specifically for TMIS. Comprising two critical phases, the DLOTDKA-MAP framework employs an Arithmetic Optimization Algorithm (AOA) and Deep Belief Network (DBN) model in real-time online threat detection, ensuring swift identification and mitigation of potential security threats. The subsequent phase employs Elliptic Curve Cryptography (ECC) with Schnorr's signature to establish secure data transmission among collaborating entities, preemptively addressing security breaches and ensuring mutual authentication. A comprehensive experimental study validates the presented DLOTDKA-MAP technique, showcasing its superiority over existing approaches and underscoring its role in advancing TMIS security infrastructure. Achieving remarkable performance values, includes an accuracy, precision, recall, and F1 score of 98.57%, this research significantly contributes to safeguarding the confidentiality and integrity of medical data within TMIS, particularly in a federated cloud environment. © The Author(s), under exclusive licence to Springer science+Business Media, LLC, part of Springer Nature 2024.

关键词： Authentication

来源：评论

学校读者我要写书评

暂无评论

Enhance the Performance of Directional Feature-based Palmprint Recognition by Directional Response Stability Measurement

引用

Machine Intelligence Research 2024年第3期21卷 597-614页

作者： Haitao Wang Wei Jia School of Computer Science and Information Engineering Hefei University of TechnologyHefei230009China

Palmprint recognition is an emerging biometrics technology that has attracted increasing attention in recent years. Many palmprint recognition methods have been proposed, including traditional methods and deep learning-based methods. Among the traditional methods, the methods based on directional features are mainstream because they have high recognition rates and are robust to illumination changes and small noises. However, to date, in these methods, the stability of the palmprint directional response has not been deeply studied. In this paper, we analyse the problem of directional response instability in palmprint recognition methods based on directional feature. We then propose a novel palmprint directional response stability measurement (DRSM) to judge the stability of the directional feature of each pixel. After filtering the palmprint image with the filter bank, we design DRSM according to the relationship between the maximum response value and other response values for each pixel. Using DRSM, we can judge those pixels with unstable directional response and use a specially designed encoding mode related to a specific method. We insert the DRSM mechanism into seven classical methods based on directional feature, and conduct many experiments on six public palmprint databases. The experimental results show that the DRSM mechanism can effectively improve the performance of these methods. In the field of palmprint recognition, this work is the first in-depth study on the stability of the palmprint directional response, so this paper has strong reference value for research on palmprint recognition methods based on directional features.

关键词： Biometrics palmprint recognition directional response stability directional coding-based methods directional feature

来源：评论

学校读者我要写书评

暂无评论

Byzantine Robust Federated Learning Scheme Based on Backdoor Triggers

引用

computers, Materials & Continua 2024年第5期79卷 2813-2831页

作者： Zheng Yang Ke Gu Yiming Zuo School of Computer and Communication Engineering Changsha University of Science and TechnologyChangsha410114China

Federated learning is widely used to solve the problem of data decentralization and can provide privacy protectionfor data owners. However, since multiple participants are required in federated learning, this allows attackers tocompromise. Byzantine attacks pose great threats to federated learning. Byzantine attackers upload maliciouslycreated local models to the server to affect the prediction performance and training speed of the global model. Todefend against Byzantine attacks, we propose a Byzantine robust federated learning scheme based on backdoortriggers. In our scheme, backdoor triggers are embedded into benign data samples, and then malicious localmodels can be identified by the server according to its validation dataset. Furthermore, we calculate the adjustmentfactors of local models according to the parameters of their final layers, which are used to defend against datapoisoning-based Byzantine attacks. To further enhance the robustness of our scheme, each localmodel is weightedand aggregated according to the number of times it is identified as malicious. Relevant experimental data showthat our scheme is effective against Byzantine attacks in both independent identically distributed (IID) and nonindependentidentically distributed (non-IID) scenarios.

关键词： Federated learning Byzantine attacks backdoor triggers

来源：评论

学校读者我要写书评

暂无评论

A Fusion Model for Personalized Adaptive Multi-Product Recommendation System Using Transfer Learning and Bi-GRU

引用

computers, Materials & Continua 2024年第12期81卷 4081-4107页

作者： Buchi Reddy Ramakantha Reddy Ramasamy Lokesh Kumar School of Computer Science and Engineering Vellore Institute of TechnologyVellore632014TamilnaduIndia

Traditional e-commerce recommendation systems often struggle with dynamic user preferences and a vast array of products,leading to suboptimal user *** address this,our study presents a Personalized Adaptive Multi-Product Recommendation System(PAMR)leveraging transfer learning and Bi-GRU(Bidirectional Gated Recurrent Units).Using a large dataset of user reviews from Amazon and Flipkart,we employ transfer learning with pre-trained models(AlexNet,GoogleNet,ResNet-50)to extract high-level attributes from product data,ensuring effective feature representation even with limited ***-GRU captures both spatial and sequential dependencies in user-item *** innovation of this study lies in the innovative feature fusion technique that combines the strengths of multiple transfer learning models,and the integration of an attention mechanism within the Bi-GRU framework to prioritize relevant *** approach addresses the classic recommendation systems that often face challenges such as cold start along with data sparsity difficulties,by utilizing robust user and item *** model demonstrated an accuracy of up to 96.9%,with precision and an F1-score of 96.2%and 96.97%,respectively,on the Amazon dataset,significantly outperforming the baselines and marking a considerable advancement over traditional *** study highlights the effectiveness of combining transfer learning with Bi-GRU for scalable and adaptive recommendation systems,providing a versatile solution for real-world applications.

关键词： Personalized recommendation systems transfer learning bidirectional gated recurrent units(Bi-GRU) performance metrics adaptive systems product reviews

来源：评论

学校读者我要写书评

暂无评论

Efficient breast cancer detection using neural networks and explainable artificial intelligence

引用

Neural Computing and Applications 2025年第5期37卷 3759-3776页

作者： Murugan, Tamilarasi Kathirvel Karthikeyan, Pritikaa Sekar, Pavithra School of Computer Science Engineering Vellore Institute of Technology Tamilnadu Chennai India

The growing dependence on deep learning models for medical diagnosis underscores the critical need for robust interpretability and transparency to instill trust and ensure responsible usage. This study investigates the efficacy of various explainable artificial intelligence (XAI) techniques in comprehending deep learning models utilized for breast cancer classification from down sampled histopathology images. A comparative assessment of multiple convolutional neural network (CNN) architectures, encompassing standard CNNs, ResNet, VGG-16, and VGG-19, on down sampled images was conducted. The primary goal is to pinpoint the model exhibiting the highest accuracy and subsequently employ three prominent XAI methods—LIME, SHAP, and Saliency Maps—to get insights into the top-performing model. This study identifies VGG-19 as the best-performing model with an accuracy of 92.59% and demonstrates that among various XAI techniques, LIME provides the most accurate and clinically relevant explanations for breast cancer classification from down sampled histopathology images. These findings, validated by medical professionals, enhance the interpretability and reliability of deep learning models in clinical settings, promoting their responsible integration into healthcare practices. This validation was further corroborated through consultation with medical professionals, including doctors specializing in breast cancer diagnosis. This research endeavors to deepen the understanding of the model’s rationale and instill confidence in its outputs. The outcomes of this study hold significant promise in elevating the interpretability and reliability of deep learning models tailored for breast cancer diagnosis, thus facilitating their responsible integration into clinical settings. © The Author(s), under exclusive licence to Springer-Verlag London Ltd., part of Springer Nature 2024.

关键词： Convolutional neural networks

来源：评论

学校读者我要写书评

暂无评论

GDMNet: A Unified Multi-Task Network for Panoptic Driving Perception

引用

computers, Materials & Continua 2024年第8期80卷 2963-2978页

作者： Yunxiang Liu Haili Ma Jianlin Zhu Qiangbo Zhang School of Computer Science and Information Engineering Shanghai Institute of TechnologyShanghai201418China

To enhance the efficiency and accuracy of environmental perception for autonomous vehicles,we propose GDMNet,a unified multi-task perception network for autonomous driving,capable of performing drivable area segmentation,lane detection,and traffic object ***,in the encoding stage,features are extracted,and Generalized Efficient Layer Aggregation Network(GELAN)is utilized to enhance feature extraction and gradient ***,in the decoding stage,specialized detection heads are designed;the drivable area segmentation head employs DySample to expand feature maps,the lane detection head merges early-stage features and processes the output through the Focal Modulation Network(FMN).Lastly,the Minimum Point Distance IoU(MPDIoU)loss function is employed to compute the matching degree between traffic object detection boxes and predicted boxes,facilitating model training *** results on the BDD100K dataset demonstrate that the proposed network achieves a drivable area segmentation mean intersection over union(mIoU)of 92.2%,lane detection accuracy and intersection over union(IoU)of 75.3%and 26.4%,respectively,and traffic object detection recall and mAP of 89.7%and 78.2%,*** detection performance surpasses that of other single-task or multi-task algorithm models.

关键词： Autonomous driving multitask learning drivable area segmentation lane detection vehicle detection

来源：评论

学校读者我要写书评

暂无评论

CALTM:A Context-Aware Long-Term Time-Series Forecasting Model

引用

computer Modeling in engineering & sciences 2024年第4期139卷 873-891页

作者： Canghong Jin Jiapeng Chen Shuyu Wu Hao Wu Shuoping Wang Jing Ying School of Computer and Computing Science Hangzhou City UniversityHangzhou310015China School of Computer Science and Engineering Macao University of Science and TechnologyMacao999078China College of Computer Science and Technology Zhejiang UniversityHangzhou310058China

Time series data plays a crucial role in intelligent transportation *** flow forecasting represents a precise estimation of future traffic flow within a specific region and time *** approaches,including sequence periodic,regression,and deep learning models,have shown promising results in short-term series ***,forecasting scenarios specifically focused on holiday traffic flow present unique challenges,such as distinct traffic patterns during vacations and the increased demand for long-term ***,the effectiveness of existing methods diminishes in such ***,we propose a novel longterm forecasting model based on scene matching and embedding fusion representation to forecast long-term holiday traffic *** model comprises three components:the similar scene matching module,responsible for extracting Similar Scene Features;the long-short term representation fusion module,which integrates scenario embeddings;and a simple fully connected layer at the head for making the final *** results on real datasets demonstrate that our model outperforms other methods,particularly in medium and long-term forecasting scenarios.

关键词： Traffic volume forecasting scene matching multi module fusion

来源：评论

学校读者我要写书评

暂无评论

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案：

请选择收藏分类：

通借通还

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：