检索结果-内蒙古大学图书馆

arXiv 2025年

作者： Liao, Haicheng Kong, Hanlin Wang, Bonan Wang, Chengyue Ye, Wang He, Zhengbing Xu, Chengzhong Li, Zhenning State Key Laboratory of Internet of Things for Smart City University of Macau China Department of Computer and Information Science University of Macau China Senseable City Lab Massachusetts Institute of Technology CambridgeMA United States

Accurate motion forecasting is crucial for safe autonomous driving (AD). This study proposes CoT-Drive, a novel approach that enhances motion forecasting by leveraging large language models (LLMs) and a chain-of-thought (CoT) prompting method. We introduce a teacher-student knowledge distillation strategy to effectively transfer LLMs’ advanced scene understanding capabilities to lightweight language models (LMs), ensuring that CoT-Drive operates in real-time on edge devices while maintaining comprehensive scene understanding and generalization capabilities. By leveraging CoT prompting techniques for LLMs without additional training, CoT-Drive generates semantic annotations that significantly improve the understanding of complex traffic environments, thereby boosting the accuracy and robustness of predictions. Additionally, we present two new scene description datasets, Highway-Text and Urban-Text, designed for fine-tuning lightweight LMs to generate context-specific semantic annotations. Comprehensive evaluations of five real-world datasets demonstrate that CoT-Drive outperforms existing models, highlighting its effectiveness and efficiency in handling complex traffic scenarios. Overall, this study is the first to consider the practical application of LLMs in this field. It pioneers the training and use of a lightweight LLM surrogate for motion forecasting, setting a new benchmark and showcasing the potential of integrating LLMs into AD systems. Copyright © 2025, The Authors. All rights reserved.

关键词： Semantics

来源：评论

学校读者我要写书评

暂无评论

Generalizable Audio Deepfake Detection via Latent Space Refinement and Augmentation

arXiv

引用

arXiv 2025年

作者： Huang, Wen Gu, Yanmei Wang, Zhiming Zhu, Huijia Qian, Yanmin Auditory Cognition and Computational Acoustics Lab MoE Key Lab of Artificial Intelligence AI Institute Department of Computer Science and Engineering Shanghai Jiao Tong University Shanghai China SJTU Paris Elite lnstitute of Technology China Ant Group Shanghai China

Advances in speech synthesis technologies, like text-to-speech (TTS) and voice conversion (VC), have made detecting deepfake speech increasingly challenging. Spoofing countermeasures often struggle to generalize effectively, particularly when faced with unseen attacks. To address this, we propose a novel strategy that integrates Latent Space Refinement (LSR) and Latent Space Augmentation (LSA) to improve the generalization of deepfake detection systems. LSR introduces multiple learnable prototypes for the spoof class, refining the latent space to better capture the intricate variations within spoofed data. LSA further diversifies spoofed data representations by applying augmentation techniques directly in the latent space, enabling the model to learn a broader range of spoofing patterns. We evaluated our approach on four representative datasets, i.e. ASVspoof 2019 LA, ASVspoof 2021 LA and DF, and In-The-Wild. The results show that LSR and LSA perform well individually, and their integration achieves competitive results, matching or surpassing current state-of-the-art methods. Copyright © 2025, The Authors. All rights reserved.

关键词： Speech recognition

来源：评论

学校读者我要写书评

暂无评论

Leveraging CNN-Transformer Hybrid Architecture for Accurate Breast Tumor Segmentation in Mammograms

Leveraging CNN-Transformer Hybrid Architecture for Accurate ...

引用

Women in Data science at Prince Sultan University (WiDS PSU), International Conference of

作者： Mudassar Ali Muhammad Zeeshan Tahir Iqra Mariam Khadija Nawaz Tariq Mahmood Samia Allaoua Chelloug College of Information Science and Electronic Engineering Zhejiang University Hangzhou China Department of Computer Science and Technology Zhejiang University Hangzhou China School of Electronics and Information Engineering Liaoning University of Technology Jinzhou China Faculty of Information Sciences University of Education Vehari Pakistan Artificial Intelligence and Data Analytics Lab CCIS PSU Riyadh SA Department of Information Technology CCIS PNU Riyadh SA

ISBN: (数字)9798331520922

ISBN: (纸本)9798331520939

Breast cancer detection through mammograms is very crucial for early diagnosis, and tumor segmentation is an important step in treatment planning. This paper proposes a novel approach by combining CNNs with transformers, which we call the Hybrid CNN + Transformer model, for the task of breast tumor segmentation. It leverages the strengths of CNN s in local feature extraction to capture fine details, such as tumor boundaries, and combines them with transformers that model long-range dependencies and contextual relationships across images for the exact delineation of tumor regions in complex tissue structures. Our methodology will be tested on the high-quality mammog-raphy dataset annotated with both benign and malignant lesions, the so-called INbreast dataset. Herein, the CNN backbone-first for instance, ResNet or EfficientNet-extracts the input mammo-gram hierarchically; features are fed into a transformer layer in order to capture global dependencies, after which the final segmentation mask is generated by using a U-Net style decoder for a pixel-wise tumor classification. The segmentation accuracies of our models demonstrated the significant improvement achieved in tumor segmentation, using Hybrid CNN + Transformer when compared with tra-ditional CNN-based models on challenging tumor boundaries and complex tissue patterns. An accurate, robust, completely automatic method for tumor segmentation will thus enable improved diagnostic feasibility on clinical grounds. Furthermore, the combination of learning at both local and global features improved generalizability to previously unseen mammogram images and therefore strengthened clinical applicability.

关键词： Image segmentation Adaptation models Accuracy Breast tumors Computational modeling Transformers Mammography Breast cancer Convolutional neural networks Tumors

来源：评论

学校读者我要写书评

暂无评论

FreeTumor: Large-Scale Generative Tumor Synthesis in Computed Tomography Images for Improving Tumor Recognition

arXiv

引用

arXiv 2025年

作者： Wu, Linshan Zhuang, Jiaxin Zhou, Yanning He, Sunan Ma, Jiabo Luo, Luyang Wang, Xi Ni, Xuefeng Zhong, Xiaoling Wu, Mingxiang Zhao, Yinghua Duan, Xiaohui Vardhanabhuti, Varut Rajpurkar, Pranav Chen, Hao Department of Computer Science and Engineering The Hong Kong University of Science and Technology Hong Kong Tencent AI Lab Shenzhen China Department of Biomedical Informatics Harvard University Boston United States Department of Computer Science and Engineering The Chinese University of Hong Kong Hong Kong Department of Radiology Shenzhen People’s Hospital Shenzhen China Department of Radiology The Third Affiliated Hospital of Southern Medical University Guangzhou China Department of Radiology Sun Yat-Sen Memorial Hospital Sun Yat-Sen University Guangzhou China Department of Diagnostic Radiology Li Ka Shing Faculty of Medicine The University of Hong Kong Hong Kong Department of Chemical and Biological Engineering The Hong Kong University of Science and Technology Hong Kong Division of Life Science The Hong Kong University of Science and Technology Hong Kong State Key Laboratory of Molecular Neuroscience The Hong Kong University of Science and Technology Hong Kong Shenzhen-Hong Kong Collaborative Innovation Research Institute The Hong Kong University of Science and Technology Shenzhen China

Tumor is a leading cause of death worldwide, with an estimated 10 million deaths attributed to tumor-related diseases every year. AI-driven tumor recognition unlocks new possibilities for more precise and intelligent tumor screening and diagnosis. However, the progress is heavily hampered by the scarcity of annotated datasets, which demands extensive annotation efforts by radiologists. To tackle this challenge, we introduce FreeTumor, an innovative Generative AI (GAI) framework to enable large-scale tumor synthesis for mitigating data scarcity. Specifically, FreeTumor effectively leverages a combination of limited labeled data and large-scale unlabeled data for tumor synthesis training. Unleashing the power of large-scale data, FreeTumor is capable of synthesizing a large number of realistic tumors on images for augmenting training datasets. To this end, we create the largest training dataset for tumor synthesis and recognition by curating 161,310 publicly available Computed Tomography (CT) volumes from 33 sources, with only 2.3% containing annotated tumors. To validate the fidelity of synthetic tumors, we engaged 13 board-certified radiologists in a Visual Turing Test to discern between synthetic and real tumors. Rigorous clinician evaluation validates the high quality of our synthetic tumors, as they achieved only 51.1% sensitivity and 60.8% accuracy in distinguishing our synthetic tumors from real ones. Through high-quality tumor synthesis, FreeTumor scales up the recognition training datasets by over 40 times, showcasing a notable superiority over state-of-the-art AI methods including various synthesis methods and foundation models. On average, FreeTumor improves the segmentation Dice scores by 6.7% and early tumor detection sensitivity by 16.4%. These findings indicate promising prospects of FreeTumor in clinical applications, potentially advancing tumor treatments and improving the survival rates of patients. Copyright © 2025, The Authors. All rights reserved.

关键词： computerized tomography

来源：评论

学校读者我要写书评

暂无评论

Canonical Shape Reconstruction with SE(3) Equivariance Learning for Weakly-Supervised Object Pose Estimation

引用

IEEE Transactions on Circuits and Systems for Video technology 2025年

作者： Zhou, Jun Chen, Kai Wei, Mingqiang Zhang, Xiao-Ping Dou, Qi Qin, Jing Hong Kong Polytechnic University Center of Smart Health School of Nursing Hong Kong Hong Kong Chinese University of Hong Kong Department of Computer Science and Engineering Hong Kong Hong Kong Nanjing University of Aeronautics and Astronautics School of Computer Science and Technology Nanjing China Tsinghua University Shenzhen Ubiquitous Data Enabling Key Lab Shenzhen International Graduate School Shenzhen China

6D object pose estimation from a single RGB-D image is a fundamental problem in computer vision and robot manipulation. Despite recent advancements, existing methods still suffer several limitations. First of all, the object shape representation extracted from the depth map is often less expressive because the object point cloud parsed from the depth map is highly incomplete due to the object self-occlusion and noisy due to the sensor artifacts. This shape representation issue further intensifies when lacking sufficient labeled data for model training, which unfortunately is another typical problem for object pose estimation considering the heavy annotation cost for real-world pose labeling. In this study, we propose to tackle the above issues in a unified way. First, we enhance the object shape representation from the partial point cloud with a novel canonical shape reconstruction module, in which an implicit canonical frame is established by incorporating the SE(3) equivariance, achieving implicit feature alignment of the partial point cloud inputs, leading to robust shape recovery. Second, based on the enhanced object representation, we further utilize the decanonicalized and pose-dependent completed object shape as the training signal, and develop a novel weakly-supervised learning framework to leverage both labeled synthetic data and unlabeled real data to train the pose estimation model in a label-efficient way. Extensive experiments on three widely used benchmarks demonstrate the effectiveness, and superiority of our framework over state-of-the-art methods. © 1991-2012 IEEE.

关键词： Self-supervised learning

来源：评论

学校读者我要写书评

暂无评论

Coded Caching Schemes for Multiaccess Topologies via Combinatorial Design

引用

IEEE Transactions on Information Theory 2025年

作者： Cheng, Minquan Wan, Kai Elia, Petros Caire, Giuseppe Guangxi Normal University Guangxi Key Lab of Multi-source Information Mining & Security Guilin541004 China Huazhong University of Science and Technology School of Electronic Information and Communications Wuhan430074 China EURECOM Communication Systems Department Sophia Antipolis06410 France Technische Universität Berlin Electrical Engineering and Computer Science Department Berlin10587 Germany

This paper studies a multiaccess coded caching (MACC) problem where the connectivity topology between the users and the caches can be described by a class of combinatorial designs. Our model includes several MACC topologies considered in previous works as special cases. The considered MACC network includes a server containing N files, Γ cache nodes and K cacheless users, where each user can access L cache nodes. The server is connected to the users via an error-free shared link, while the users can directly access the content in their connected cache nodes. Our goal is to minimize the worst-case transmission load on the shared link in the delivery phase. The main limitation of the existing MACC works is that only some specific access topologies are considered, including the only cases where the number of users K scales either linearly or exponentially in Γ. We overcome this limitation by formulating a new access topology derived from two classical combinatorial structures, referred to as the t-design and the t-group divisible design. By leveraging the properties of these combinatorial structures, we propose two classes of coded caching schemes for a flexible number of users, where the number of users can scale linearly, polynomially or exponentially with the number of cache nodes. As a by-product, by extending the proposed scheme to the original dedicated coded caching scenario (i.e., each user has its own cache), the resulting scheme can unify several existing coded caching schemes. © 1963-2012 IEEE.

关键词： Topology

来源：评论

学校读者我要写书评

暂无评论

Metaverse meets distributed machine learning: A contemporary review on the development with privacy-preserving concerns

引用

ICT Express 2025年

作者： Gardner, Samuel Harry Hoang, Trong-Minh Na, Woongsoo Dao, Nhu-Ngoc Cho, Sungrae Department of Computer Science and Engineering Sejong University Seoul 05006 South Korea Intelligently Connected Networks Lab Posts and Telecommunications Institute of Technology Hanoi 100000 Viet Nam Division of Computer Science and Engineering Kongju National University Cheonan 31080 South Korea School of Computer Science and Engineering Chung-Ang University Seoul 05006 South Korea

Distributed machine learning utilization in the metaverse exposes many potential benefits. However, the combination of these advanced technologies raises significant privacy concerns due to the potential exploitation of sensitive user and system data. This paper provides a systematic investigation of over 100 recent studies across key academic databases obtained by initial keyword-filter screening followed by a thorough full-text review. Particularly, metaverse evolution and enabling infrastructure technologies are briefly summarized. Subsequently, the distributed learning architectures and their features are analyzed as well as possibly associated vulnerability discussions. Then, envisioned metaverse applications and future research challenges are highlighted before concluding remarks. © 2025 The Authors

关键词： Distributed machine learning Mobile metaverse Privacy preservation Virtual reality

来源：评论

学校读者我要写书评

暂无评论

NavG: Risk-Aware Navigation in Crowded Environments Based on Reinforcement Learning with Guidance Points

arXiv

引用

arXiv 2025年

作者： Zhang, Qianyi Luo, Wentao Liu, Boyi Zhang, Ziyang Wang, Yaoyuan Liu, Jingtai Institute of Robotics and Automatic Information System Nankai University Tianjin Key Laboratory of Intelligent Robotics Tianjin China Advanced Computing and Storage Lab Huawei 2012 Lab China Department of Electronic and Computer Engineering The Kong Kong University of Science and Technology Hong Kong

Motion planning in navigation systems is highly susceptible to upstream perceptual errors, particularly in human detection and tracking. To mitigate this issue, the concept of guidance points—a novel directional cue within a reinforcement learning-based framework—is introduced. A structured method for identifying guidance points is developed, consisting of obstacle boundary extraction, potential guidance point detection, and redundancy elimination. To integrate guidance points into the navigation pipeline, a perception-to-planning mapping strategy is proposed, unifying guidance points with other perceptual inputs and enabling the RL agent to effectively leverage the complementary relationships among raw laser data, human detection and tracking, and guidance points. Qualitative and quantitative simulations demonstrate that the proposed approach achieves the highest success rate and near-optimal travel times, greatly improving both safety and efficiency. Furthermore, real-world experiments in dynamic corridors and lobbies validate the robot’s ability to confidently navigate around obstacles and robustly avoid pedestrians. © 2025, CC BY.

关键词： Travel time

来源：评论

学校读者我要写书评

暂无评论

EXCEPTION HANDLING - EXPECTING THE UNEXPECTED

引用

computer LANGUAGES 1994年第2期20卷 69-87页

作者： DREW, SJ GOUGH, KJ Programming Languages and Systems Group Computer Science Department Queensland University of Technology Queensland Australia 4000

Since the mid-1970s, and with the development of each new programming paradigm there has been an increasing interest in exceptions and the benefits of exception handling. With the move towards programming for ever more complex architectures, understanding basic facilities such as exception handling as an aid to improving program reliability, robustness and comprehensibility has become much more important. Interest has sparked the production of many papers both theoretical and practical, each giving a view of exceptions and exception handling from a different standpoint. In an effort to provide a means of classifying exception handling models which may be encountered, a taxonomy is presented in this paper. As the taxonomy is developed some of the concepts of exception handling are introduced and discussed. The taxonomy is applied to a number of exception handling models in some contemporary programming languages and some observations and conclusions offered.

关键词： EXCEPTIONS EXCEPTION HANDLING programming LANGUAGES

来源：评论

学校读者我要写书评

暂无评论

Block Sparse Vector Codes for Ultra-Reliable and Low-Latency Short-Packet Transmission

引用

IEEE Transactions on Communications 2025年

作者： Zhang, Yanfeng Zhu, Xu Liu, Yujie Fan, Xi'an Guan, Yong Liang Wong, M. L. Dennis Lau, Vincent K. N. Dongguan University of Technology School of Electrical Engineering and Intelligentization Dongguan China Harbin Institute of Technology School of Information Science and Technology Shenzhen518055 China Guangxi Normal University Guangxi Key Laboratory of Brain Inspired Computing and Intelligent Chips Guilin541001 China Shenzhen Municipal Key Laboratory of AIoT Communications Shenzhen518055 China Nanyang Technological University Continental-NTU Corporate Lab Singapore Dongguan University of Technology School of Computer Science and Technology Dongguan China Newcastle University School of Engineering Newcastle upon TyneNE1 7RU United Kingdom The Hong Kong University of Science and Technology Department of Electronic and Computer Engineering Hong Kong

Sparse vector coding (SVC) is a promising short-packet transmission method for ultra reliable low latency communication (URLLC) in next-generation communication systems. In this paper, a block SVC (BSVC) based short-packet transmission scheme is proposed to further enhance the transmission performance of SVC. The core idea behind the proposed scheme is to transmit short-packet data information after block sparse transformation. At the transmitter, the information bits are divided into two parts: one part is mapped into the non-zero indexes of block sparse vectors, and the other part is mapped into non-zero values through phase or amplitude modulation. After pseudo-random spreading, the block sparse vectors are mapped to time-frequency resources for transmission. At the receiver, the decoding problem is transformed into a block sparse signal recovery problem. A cyclic block orthogonal matching pursuit (CBOMP) algorithm is proposed for decoding by leveraging block-structured sparse prior information. The upper bound of block error rate (BLER) performance over Rayleigh channels is derived to verify the decoding performance of the proposed CBOMP algorithm. Extensive simulation results verify that the proposed BSVC scheme outperforms the existing SVC schemes in terms of BLER, transmission latency and spectral efficiency over fading channels. © 1972-2012 IEEE.

关键词： Fading channels

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：