检索结果-内蒙古大学图书馆

SSRN 2024年

作者： Liang, Mengnan Wang, Congcong Zhao, Meng Shi, Fan Zhang, Chao Han, Yilin School of Computer Science and Engineering Tianjin University of Technology Tianjin300384 China Laboratory of Computer Vision System of Ministry of Education Tianjin University of Technology Tianjin300384 China

Monocular depth estimation is a classic research topic in computer vision. In recent years, development of Convolutional Neural Networks (CNNs) has facilitated significant breakthroughs in this field. However, there still are two challenges: 1) The network struggles to effectively fuse edge features in the feature fusion stage, which ultimately results in the loss of structure or distorted boundaries of objects in the scene. 2) Classification based studies typically depend on Transformers for global modeling, a process that often introduces substantial performance overhead. In this paper, we propose two modules to address the aforementioned issues. The first module is the Boundary Attention Module (BAM), which leverages the attention mechanism to enhance the ability of the network to perceive object boundaries during the feature fusion stage. In addition, to mitigate the performance overhead resulting from predicting adaptive bins, we propose a Shift Window Adaptive Bins (SWAB) module to reduce the amount of calculation in global modeling. The proposed method is evaluated on two public datasets, NYU Depth V2 and KITTI, and achieves state-of-the-art (SOTA) performance. © 2024, The Authors. All rights reserved.

关键词： computer vision

来源：评论

学校读者我要写书评

暂无评论

How to Select Pre-Trained Code Models for Reuse? A Learning Perspective

arXiv

引用

arXiv 2025年

作者： Bi, Zhangqian Wan, Yao Chu, Zhaoyang Hu, Yufei Zhang, Junyi Zhang, Hongyu Xu, Guandong Jin, Hai National Engineering Research Center for Big Data Technology and System Services Computing Technology and System Lab Cluster and Grid Computing Lab Wuhan China School of Computer Science and Technology Huazhong University of Science and Technology Wuhan China School of Big Data and Software Engineering Chongqing University Chongqing China School of Computer Science University of Technology Sydney Sydney Australia

Pre-training a language model and then fine-tuning it has shown to be an efficient and effective technique for a wide range of code intelligence tasks, such as code generation, code summarization, and vulnerability detection. However, pre-training language models on a large-scale code corpus is computationally expensive. Fortunately, many off-the-shelf Pre-trained Code Models (PCMs), such as CodeBERT, CodeT5, CodeGen, and Code Llama, have been released publicly. These models acquire general code understanding and generation capability during pre-training, which enhances their performance on downstream code intelligence tasks. With an increasing number of these public pre-trained models, selecting the most suitable one to reuse for a specific task is essential. In this paper, we systematically investigate the reusability of PCMs. We first explore three intuitive model selection methods that select by size, training data, or brute-force fine-tuning. Experimental results show that these straightforward techniques either perform poorly or suffer high costs. Motivated by these findings, we explore learning-based model selection strategies that utilize pre-trained models without altering their parameters. Specifically, we train proxy models to gauge the performance of pre-trained models, and measure the distribution deviation between a model's latent features and the task's labels, using their closeness as an indicator of model transferability. We conduct experiments on 100 widely-used open-source PCMs for code intelligence tasks, with sizes ranging from 42.5 million to 3 billion parameters. The results demonstrate that learning-based selection methods reduce selection time to 100 seconds, compared to 2,700 hours with brute-force fine-tuning, with less than 6% performance degradation across related tasks. © 2025, CC BY-NC-SA.

关键词： Adversarial machine learning

来源：评论

学校读者我要写书评

暂无评论

Invertible Mosaic Image Hiding Network for Very Large Capacity Image Steganography

Invertible Mosaic Image Hiding Network for Very Large Capaci...

引用

International Conference on Acoustics, Speech, and Signal Processing (ICASSP)

作者： Zihan Chen Tianrui Liu Jun-Jie Huang Wentao Zhao Xing Bi Meng Wang College of Computer Science and Technology National University of Defense Technology China College of System Engineering National University of Defense Technology China School of Computer Science and Information Engineering Hefei University of Technology China

The existing image steganography methods either sequentially conceal secret images or conceal a concatenation of multiple images. In such ways, the interference of information among multiple images will become increasingly severe when the number of secret images becomes larger, thus restrict the development of very large capacity image steganography. In this paper, we propose an Invertible Mosaic Image Hiding Network (InvMIHNet) which realizes very large capacity image steganography with high quality by concealing a single mosaic secret image. InvMIHNet consists of an Invertible Image Rescaling (IIR) module and an Invertible Image Hiding (IIH) module. The IIR module works for downscaling the single mosaic secret image form by spatially splicing the multiple secret images, and the IIH module then conceal this mosaic image under the cover image. The proposed InvMIHNet successfully conceal and reveal up to 16 secret images with a small number of parameters and memory consumption. Extensive experiments on ImageNet-1K, COCO and DIV2K show InvMIHNet outperforms state-of-the-art methods in terms of both the imperceptibility of stego image, recover accuracy of secret image and security against steganlysis methods. The code is available at https://***/Brittany-Chen/InvMIHNet.

关键词：

来源：评论

学校读者我要写书评

暂无评论

DEGSTalk: Decomposed Per-Embedding Gaussian Fields for Hair-Preserving Talking Face Synthesis

DEGSTalk: Decomposed Per-Embedding Gaussian Fields for Hair-...

引用

International Conference on Acoustics, Speech, and Signal Processing (ICASSP)

作者： Kaijun Deng Dezhi Zheng Jindong Xie Jinbao Wang Weicheng Xie Linlin Shen Siyang Song Computer Vision Institute School of Computer Science and Software Engineering Shenzhen University National Engineering Laboratory for Big Data System Computing Technology Shenzhen University Guangdong Provincial Key Laboratory of Intelligent Information Processing Department of Computer Science University of Exeter

ISBN: (数字)9798350368741

ISBN: (纸本)9798350368758

Accurately synthesizing talking face videos and capturing fine facial features for individuals with long hair presents a significant challenge. To tackle these challenges in existing methods, we propose a decomposed per-embedding Gaussian fields (DEGSTalk), a 3D Gaussian Splatting (3DGS)-based talking face synthesis method for generating realistic talking faces with long hairs. Our DEGSTalk employs Deformable Pre-Embedding Gaussian Fields, which dynamically adjust pre-embedding Gaussian primitives using implicit expression coefficients. This enables precise capture of dynamic facial regions and subtle expressions. Additionally, we propose a Dynamic Hair-Preserving Portrait Rendering technique to enhance the realism of long hair motions in the synthesized videos. Results show that DEGSTalk achieves improved realism and synthesis quality compared to existing approaches, particularly in handling complex facial dynamics and hair preservation. Our code is available at https://***/CVI-SZU/DEGSTalk.

关键词： Hair Training Three-dimensional displays Dynamics Signal processing Rendering (computer graphics) Noise measurement Speech processing Faces Videos

来源：评论

学校读者我要写书评

暂无评论

Multi-secret Sharing Scheme for Zero Trust Environment 14

Multi-secret Sharing Scheme for Zero Trust Environment

引用

14th IEEE International Conference on Software engineering and Service Science, ICSESS 2023

作者： Zhou, Qin Yang, Rumiao Zhou, Hao Cai, Lei Yongyou Network Technology Co. Ltd Beijing China School of Computer and Communication Engineering University of Science and Technology Beijing Beijing China North China Institute of Computing Technology The 4th System Department Beijing China

ISBN: (纸本)9798350336269

The zero-trust concept of 'never trust, always verify' injected new vitality into the development of network security. However, the related research on zero trust is still preliminary, and the research on secure data transmission is relatively less. How to achieve secure data transmission in zero trust environment is a problem to be solved. Aiming the data transmission scenario of private data secret sharing among agents, this paper proposed a secure data transmission scheme based on multi-secret sharing to solve the problems of identity comparison confirmation of participants and adjust the access structure according to the security value of trust evaluation in existing secret sharing schemes. The proposed scheme adopts the secret sharing design without the trusted third party, which avoids the security risk of trusted third party disclosure in the traditional secret sharing scheme. Secondly, in the system parameter setting phase, the identity of participants and zero-trust identity database are compared to combine the characteristics of zero-trust 'identity authentication', to improve the security of secret sharing. Finally, during secret distribution, zero-trust trust evaluation security values are used to adjust the access structure based on zero-trust 'trust evaluation' to ensure secure data transmission. © 2023 IEEE.

关键词： Data communication systems

来源：评论

学校读者我要写书评

暂无评论

How to Select Pre-Trained Code Models for Reuse? A Learning Perspective

How to Select Pre-Trained Code Models for Reuse? A Learning ...

引用

IEEE International Conference on Software Analysis, Evolution and Reengineering (SANER)

作者： Zhangqian Bi Yao Wan Zhaoyang Chu Yufei Hu Junyi Zhang Hongyu Zhang Guandong Xu Hai Jin National Engineering Research Center for Big Data Technology and System Services Computing Technology and System Lab Cluster and Grid Computing Lab Wuhan China School of Computer Science and Technology Huazhong University of Science and Technology Wuhan China School of Big Data and Software Engineering Chongqing University Chongqing China School of Computer Science University of Technology Sydney Sydney Australia

ISBN: (数字)9798331535100

ISBN: (纸本)9798331535117

Pre-training a language model and then fine-tuning it has shown to be an efficient and effective technique for a wide range of code intelligence tasks, such as code generation, code summarization, and vulnerability detection. However, pre-training language models on a large-scale code corpus is compu-tationally expensive. Fortunately, many off-the-shelf Pre-trained Code Models (PCMs), such as CodeBERT, CodeT5, CodeGen, and Code Llama, have been released publicly. These models acquire general code understanding and generation capability during pre-training, which enhances their performance on downstream code intelligence tasks. With an increasing number of these public pre-trained models, selecting the most suitable one to reuse for a specific task is essential. In this paper, we systematically investigate the reusability of PCMs. We first explore three intuitive model selection methods that select by size, training data, or brute-force fine-tuning. Experimental results show that these straightforward techniques either perform poorly or suffer high costs. Motivated by these findings, we explore learning-based model selection strategies that utilize pre-trained models without altering their parameters. Specifically, we train proxy models to gauge the performance of pre-trained models, and measure the distribution deviation between a model's latent features and the task's labels, using their closeness as an indicator of model transferability. We conduct experiments on 100 widely-used open-source PCMs for code intelligence tasks, with sizes ranging from 42.5 million to 3 billion parameters. The results demonstrate that learning-based selection methods reduce selection time to 100 seconds, compared to 2,700 hours with brute-force fine-tuning, with less than 6% performance degradation across related tasks.

关键词： Phase change materials Analytical models Adaptation models Codes Costs Computational modeling Training data Machine learning Software Software development management

来源：评论

学校读者我要写书评

暂无评论

Fast Foothold Planning for Quadruped Robot Locomotion Using Topological Perception

Fast Foothold Planning for Quadruped Robot Locomotion Using ...

引用

International Electronics Symposium (IES)

作者： Azhar Aulia Saputra Masaya Shoji Fernando Ardilla Naoyuki Kubota Department of Mechanical Systems Engineering Graduate School of System Design Tokyo Metropolitan University Japan Department of Informatics and Computer Engineering Politeknik Elektronika Negeri Surabaya Indonesia

ISBN: (数字)9798350391992

ISBN: (纸本)9798350392005

Recent advancements in legged robot locomotion systems have led to significant improvements in their agility and overall performance. However, for these systems to effectively handle sudden obstacles, especially in foothold areas, real-time integration of external information is crucial. In this paper, we introduce a rapid integration approach between perception and action in legged robot locomotion to address such unexpected external obstacles. We utilize depth sensors to capture external data in the form of pointclouds. We have developed a fast surface perception model based on topological information, employing dynamic sparsity through the growing neural gas algorithm. This model addresses several key challenges in 3D point cloud extraction: 1) Extracting object features from a smaller portion of data. 2) Controlling granularity in unsupervised learning using a topological structure. 3) Controlling granularity with only 3D positional point cloud information. To evaluate the effectiveness of our proposed method, we conducted a series of progressive experiments. Initially, we compared the performance of our topological perception model with similar approaches. Subsequently, we carried out qualitative and quantitative experiments using a real, low-cost quadruped robot

关键词： Point cloud compression Legged locomotion Solid modeling Three-dimensional displays Heuristic algorithms Feature extraction Sensors

来源：评论

学校读者我要写书评

暂无评论

A Speed Measurement Method of Rail Transit Based on Millimeter-Wave Radar 1

引用

5th International Symposium for Intelligent Transportation and Smart City, ITASC 2022

作者： Shen, Tuo Xie, Lanxin Luo, Jinnan Zeng, Xiaoqing Shanghai Key Laboratory of Rail Infrastructure Durability and System Safety Shanghai201804 China School of Optical-Electrical and Computer Engineering University of Shanghai for Science and Technology Shanghai200093 China Institute of Rail Transit Tongji University Shanghai China

ISBN: (数字)9789819922529

ISBN: (纸本)9789819922512

The growing use of millimeter-wave radars in the intelligent transportation field and increasing accuracy provide new possibilities for train speed measurement. This paper proposes a measuring train speed method by using millimeter-wave radar to detect static objects. The millimeter-wave radar was used to obtain and process objects data. Then, using the RANSAC algorithm to extract the static objects group, and calculated the relative speed of the millimeter-wave radar and the static objects group. Finally, we completed the comparative experiment of using millimeter-wave radar and odometer to measure train speed in a real track environment. The results show that the measurement error of this method is less than 1.5 m/s, and it has the characteristics of low cost, fast speed measurement, and high precision. © 2023, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.

关键词： Object detection

来源：评论

学校读者我要写书评

暂无评论

Joint Resource Allocation and Blocklength Assignment in Star-Ris and Noma-Assisted Urllc systems

SSRN

引用

SSRN 2024年

作者： Yang, Shuhui Chen, Jian Jie, Jia Guo, Liang Wang, Xingwei School of Computer Science and Engineering Northeastern University Liaoning Shenyang110819 China Engineering Research Center of Security Technology of Complex Network System Ministry of Education China

This paper investigates a novel non-orthogonal multiple access (NOMA) assisted ultra-reliable low-latency communications (URLLC) paradigm. By allowing multiple users to occupy the same time or frequency resource blocks with the NOMA technique, the transmission delays and network congestion in multi-user URLLC communication systems are thus mitigated. Furthermore, the simultaneously transmitting and reflecting reconfigurable intelligent surface (STAR-RIS) is employed to provide full-space coverage by proactively reconfiguring the electromagnetic propagation environment. The energy consumption optimization problem is formulated by refining blocklength assignment and jointly optimizing the power allocation and the STAR-RIS configuration. Due to highly coupled variables and the non-convexity of constraints, the optimization problem is proven to be NP-hard. To tackle the intractable problem, we propose an alternating optimization (AO)-based solution combining adaptive genetic algorithm (AGA), and successive convex approximation (SCA) methods. Numerical results demonstrate that: i) the proposed refined blocklength assignment scheme outperforms that with fixed blocklength assignment;ii) the proposed STAR-RIS-assisted NOMA system is superior to the conventional RIS-assisted NOMA system. © 2024, The Authors. All rights reserved.

关键词： Resource allocation

来源：评论

学校读者我要写书评

暂无评论

A Robust Anomaly Detection Framework Utilizing Feature engineering and Ensemble Techniques with BETH Data

A Robust Anomaly Detection Framework Utilizing Feature Engin...

引用

2024 International Conference on Intelligent systems for Cybersecurity, ISCS 2024

作者： Neeraja, Kondamudi Naga Rao, T. Prabhakara Rao, Peluru Janardhana Pbv, Raja Rao Moses, G. Jose Lakshmanarao, A. Andhra University College of Engineering Au Department of Computer Science and System's Engineering Visakhapatnam India Aditya Unviersity Department of Cse Surampalem India Avanthi's Research Technological Academy Department of Cse A.P Vizianagaram India Department of Cse A.P Bhimavaram India School of Engineering Malla Reddy University Hyderabad India Aditya Unviersity Department of It Surampalem India

ISBN: (纸本)9798350375237

Anomaly detection holds a critical role in cybersecurity, with the aim of identifying suspicious activities or behaviors that deviated from normal patterns. In this paper, the BPF-extended tracking honeypot (BETH) dataset is utilized, designed specifically for uncertainty and robustness benchmarking in cybersecurity. BETH consists of essential features and the target variable evil. This dataset captured modern host activity and data collected from cloud services, rendering it highly relevant for real-world deployments. Notably, BETH offered a comprehensive view of malicious behavior by incorporating both kernel-process and network logs. The study aimed to apply various machine learning algorithms, encompassing Adaboost, XGBoost, Gradient Boosting, Random Forest, and Extra Trees Classifier, to detect anomalies in the BETH dataset. Prior to model training, preprocessing techniques were employed to handle missing values, perform feature engineering, and scale numerical features. Additionally, the performance of these algorithms was compared, and exploration into ensemble techniques stacking and voting classifiers to enhance detection accuracy. Through systematic experimentation and evaluation, the goal was to identify the most effective anomaly detection algorithms and ensemble strategies for enhancing cybersecurity measures. © 2024 IEEE.

关键词： Anomaly detection

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：