检索结果-内蒙古大学图书馆

22nd IEEE International Symposium on Parallel and Distributed Processing with Applications, ISPA 2024

作者： Tu, Ping-Han Cheng, Yu-Che Hong, Ding-Yong Liu, Pangfeng Wu, Jan-Jan Institute of Information Science Academia Sinica Taipei Taiwan National Taiwan University Department of Computer Science and Information Engineering Taipei Taiwan National Taiwan University The Graduate Institute of Networking and Multimedia Taipei Taiwan

ISBN: (纸本)9798331509712

Convolutional neural networks (CNNs) have achieved immense success in computer vision and other field of science. Despite the achievements, state-of-the-art CNN models have grown to gigantic sizes that demand a lot of computing power and memory resources. As a model compression technique, parameter pruning can lower the computation and memory a CNN model needs. Previous work has proposed a parameter pruning method, which compresses the neural networks by eliminating the rows and columns of the weight matrices. Removing rows and columns preserves the dense structure of the weight matrix, rather than the sparse structure produced by unstructured pruning. Although row-and-column pruning effectively reduces the model size, making the model as compact as possible without sacrificing accuracy remains a challenge. We prove that the row-and-column pruning problem is an NP-complete problem, and we propose two approximation algorithms that provide solutions within twice the cost of the optimal solution. We also show that the approximation ratio cannot be improved for these two algorithms. Finally, we propose two schemes based on simulated annealing to solve the row-and-column pruning problem. The two proposed schemes improve accuracy by 2.7% and 1.6%, respectively, on the ResNet-18 model with 93% sparsity using the Food-101 dataset. © 2024 IEEE.

关键词： Convolutional neural networks

来源：评论

学校读者我要写书评

暂无评论

Deep Learning Based End-to-End Specular Reflection Removal for Medical Endoscopic Images 23

Deep Learning Based End-to-End Specular Reflection Removal f...

引用

2023 Research in Adaptive and Convergent Systems, RACS 2023

作者： Shih, Chi-Sheng Liao, Yu-Cheng Tan, Ching-Ting Graduate Institute of Networking and Multimedia Department of Computer Science and Information Engineering High Performance and Scientific Computing Center National Taiwan University Taipei Taiwan Graduate Institute of Networking and Multimedia National Taiwan University Taipei Taiwan Department of Otolaryngology National Taiwan University Hospital Taipei Taiwan

ISBN: (纸本)9798400702280

Endoscopic images usually have many overexposed regions due to strong and focused light sources and, consequently, physicians need to change the camera angle for a clear view from time to time. This work targets removing the specular reflections from the endoscopy images but maintaining the structural and semantic texture in the images. Unlike the state-of-the-art image inpainting methods, the proposed method treats the specular reflection removal problem as an image-to-image translation task by skipping the specular detection step. The technique uses Dense Multi-scale Fusion Network (DMFN), which aims for small size and clear images, as the base and revises the network to support the large size and noisy endoscopy images. Moreover, the method can recover structure from large overexposed regions. Experimental results show that under different glare ratios and granularity settings, it is more effective than the Partial Convolution Model when recovering the corrupted pixels: the mean average error of pixel values decreases 13.2%, and the PSNR value increases 2.62dB. Qualitative analysis shows that even in the worst case when the proposed model generates output images with MAE up to 4.30, its recoveries still possess good visual qualities. The proposed model can process a 480 × 768 image in 16.2ms on a computer equipped with 36 CPU cores and an RTX-3090 GPU. It usually takes at least 20 minutes for a human to recover such images. © 2023 ACM.

关键词： Endoscopy

来源：评论

学校读者我要写书评

暂无评论

Paths Optimization by Jointing Link Management and Channel Estimation Using Variational Autoencoder With Attention for IRS-MIMO Systems

IEEE Transactions on Machine Learning in Communications and ...

引用

IEEE Transactions on Machine Learning in Communications and networking 2025年 3卷 381-394页

作者： Wu, Meng-Hsun Chen, Hong-Yunn Yang, Ta-Wei Hsu, Chih-Chuan Huang, Chih-Wei Chou, Cheng-Fu National Taiwan University Department of Computer Science and Information Engineering Taipei10617 Taiwan National Pingtung University Department of Computer Science and Information Engineering Pingtung91201 Taiwan Graduate Institute of Networking and Multimedia National Taiwan University Taipei10617 Taiwan Texas A&M University Department of Computer Science and Engineering College StationTX77840 United States National Central University Department of Communication Engineering Taoyuan320 Taiwan

In massive MIMO systems, achieving optimal end-to-end transmission encompasses various aspects such as power control, modulation schemes, path selection, and accurate channel estimation. Nonetheless, optimizing resource allocation remains a significant challenge. In path selection, the direct link is a straightforward link between the transmitter and the receiver. On the other hand, the indirect link involves reflections, diffraction, or scattering, often due to interactions with objects or obstacles. Relying exclusively on one type of link can lead to suboptimal and limited performance. Link management (LM) is emerging as a viable solution, and accurate channel estimation provides essential information to make informed decisions about transmission parameters. In this paper, we study LM and channel estimation that flexibly adjust the transmission ratio of direct and indirect links to improve generalization, using a denoising variational autoencoder with attention modules (DVAE-ATT) to enhance sum rate. Our experiments show significant improvements in IRS-assisted millimeter-wave MIMO systems. Incorporating LM increased the sum rate and reduced MSE by approximately 9%. Variational autoencoders (VAE) outperformed traditional autoencoders in the spatial domain, as confirmed by heatmap analysis. Additionally, our investigation of DVAE-ATT reveals notable differences in the temporal domain with and without attention mechanisms. Finally, we analyze performance across varying numbers of users and ranges. Across various distances—5m, 15m, 25m, and 35m—performance improvements averaged 6%, 11%, 16%, and 22%, respectively. © 2023 CCBY.

关键词： Resource allocation

来源：评论

学校读者我要写书评

暂无评论

Attention-Aided Autoencoder-Based Channel Prediction for Intelligent Reflecting Surface-Assisted Millimeter-Wave Communications

IEEE Transactions on Green Communications and Networking

引用

IEEE Transactions on Green Communications and networking 2023年第4期7卷 1906-1919页

作者： Chen, Hong-Yunn Wu, Meng-Hsun Yang, Ta-Wei Huang, Chih-Wei Chou, Cheng-Fu National Taiwan University Graduate Institute of Networking and Multimedia Taipei10617 Taiwan National Taiwan University Department of Computer Science and Information Engineering Taipei10617 Taiwan National Central University Department of Communication Engineering Taoyuan320317 Taiwan

Sixth-generation (6G) wireless communication networks will provide larger coverage and capacity with lower energy consumption and hardware costs than 5G. Intelligent reflecting surface (IRS)-aided millimeter-wave massive MIMO OFDM communication is a new technology that intelligently manipulates electromagnetic waves. This has recently attracted much attention given its potential to manage the wireless propagation environment at low hardware costs and with minimal energy usage. However, channel prediction is complicated by the fact that IRS is rarely equipped with power amplifiers, various radio frequency chains, or a significant number of reflecting components. In this paper, we propose a convolutional denoising autoencoder model and investigate a joint attention mechanism for channel prediction. Then, we employ the attention mechanism to identify features of channel subcarrier interference to improve the channel prediction performance. Long-range dependent specificity is captured through the attention mechanism to generate useful features from the input signal. The encoder-decoder design of the autoencoder serves as a dimensionality reduction method that enables the autoencoder to predict the spatial and temporal distribution features of continuous signals by exploiting the extraction of sequence features from the model. Numerical results show that the proposed algorithm significantly improves the performance of IRS-aided millimeter-wave massive MIMO OFDM communication systems compared with previous methods. © 2017 IEEE.

关键词： Orthogonal frequency division multiplexing

来源：评论

学校读者我要写书评

暂无评论

Singer Separation for Karaoke Content Generation 27

Singer Separation for Karaoke Content Generation

引用

27th Conference on the Oriental COCOSDA International Committee for the Co-Ordination and Standardisation of Speech Databases and Assessment Techniques, O-COCOSDA 2024

作者： Lin, Hsuan-Yu Chen, Xuanjun Jang, Jyh-Shing Roger Graduate Institute of Networking and Multimedia National Taiwan University Taipei Taiwan Graduate Institute of Communication Engineering National Taiwan University Taipei Taiwan National Taiwan University Dept. of Computer Science and Information Engineering Taipei Taiwan

ISBN: (纸本)9798331506032

Due to the rapid development of deep learning, we can now successfully separate singing voice from mono audio music. However, this separation can only extract human voices from other musical instruments, which is undesirable for karaoke content generation applications that only require the separation of lead singers. For this karaoke application, we need to separate the music containing male and female duets into two vocals, or extract a single lead vocal from the music containing vocal harmony. For this reason, we propose in this article to use a singer separation system, which generates karaoke content for one or two separated lead singers. In particular, we introduced three models for the singer separation task and designed an automatic model selection scheme to distinguish how many lead singers are in the song. We also collected a large enough data set, MIR-SingerSeparation11, which has been publicly released to advance the frontier of this research. Our singer separation is most suitable for sentimental ballads and can be directly applied to karaoke content generation. As far as we know, this is the first singer-separation work for real-world karaoke applications. © 2024 IEEE.

关键词： Source separation

来源：评论

学校读者我要写书评

暂无评论

Function Clustering to Optimize Resource Utilization on Container Platform 29

Function Clustering to Optimize Resource Utilization on Cont...

引用

29th IEEE International Conference on Parallel and Distributed Systems, ICPADS 2023

作者： Lee, Chao-Yu Hong, Ding-Yong Liu, Pangfeng Wu, Jan-Jan Academia Sinica Institute of Information Science Taipei Taiwan National Taiwan University Department of Computer Science and Information Engineering Taipei Taiwan National Taiwan University The Graduate Institute of Networking and Multimedia Taipei Taiwan

ISBN: (纸本)9798350330717

In recent years, container technology has gained significant attention in the software industry, with many businesses opting for its elasticity, cost-effectiveness, and ease of implementation. The most common way to ensure a seamless user experience is to keep a substantial number of containers active throughout the day, which causes resource over-provision. Conversely, closing the container right after handling the requests can reduce memory consumption but generate a cold start whenever the request arrives. Cold start occurrence and resource usage is a trade-off and presents a significant challenge on the container *** address this challenge, we observe that serving consecutive requests with the same container can notably decrease the number of cold starts. We propose TAC, a Temporal Adjacency Function Clustering algorithm, to meet the challenge. TAC selects the functions with time adjacency requests into a cluster from the historical data. TAC packs functions serving time adjacency requests into a cluster to reduce cold starts and enable efficient resource utilization. The experiment result shows that TAC reduces 8% cold start occurrences and 53% memory usage with the real-world traces compared to the state-of-the-art methods, e.g., Defuse [1] and Hybrid histogram policy [2]. © 2023 IEEE.

关键词： cold-start container platform Function-as-a-Service serverless

来源：评论

学校读者我要写书评

暂无评论

Breathero: Not Another Slow Breathing Game - Exploring Faster-Paced VR Breathing Exercise Games

Breathero: Not Another Slow Breathing Game - Exploring Faste...

引用

Extended Abstracts of the 2023 CHI Conference on Human Factors in Computing Systems, CHI EA 2023

作者： Wang, Shun-Yu Cheng, Chia-Yu Huang, Shu-Meng Chan, Weng Io Huang, Yu-Hsuan Lin, Jhih-Wei >National Taiwan University Graduate Institute of Networking and Multimedia Taiwan National Taiwan University Department of Computer Science and Information Engineering Taiwan National Taiwan University of Science and Technology Department of Design Taiwan

ISBN: (纸本)9781450394222

BreatHero is a faster-paced Virtual Reality (VR) action game that explores the potential of combining breathing techniques learning with the combat elements of action games. BreatHero uses breathing techniques as various skills for attacking and defending against the enemies, so players can do breathing exercises while enjoying defeating the in-game enemies. We choose three breathing techniques: Kapalabhati, Box Breathing, and Full Yogic Breathing, which are all beneficial for physical and mental health. To enhance breathing techniques learning and memorization in action games, we present two design concepts: Imitative Breathing Feedback and Associate Learning on Breathing Techniques. Results from our study indicated that players found the game to be more engaging and were willing to play it repeatedly, suggesting that breathing exercises can be effectively integrated into faster-paced action games. © 2023 Owner/Author.

关键词： Virtual reality

来源：评论

学校读者我要写书评

暂无评论

Accelerate Inference of CNN Models on CPU via Column Combining Based on Simulated Annealing 11

Accelerate Inference of CNN Models on CPU via Column Combini...

引用

11th International Symposium on Computing and networking, CANDAR 2023

作者： Lin, Chien-Hung Hong, Ding-Yong Liu, Pangfeng Wu, Jan-Jan Institute of Information Science Academia Sinica Taipei Taiwan National Taiwan University Department of Computer Science and Information Engineering Taipei Taiwan The Graduate Institute of Networking and Multimedia National Taiwan University Taipei Taiwan National Taiwan University The Graduate School of Advanced Technology Taipei Taiwan

ISBN: (纸本)9798350306705

Convolutional Neural Networks (CNNs) have been successful in various computer vision tasks. However, the size of state-of-The-Art CNN models tends to be tremendous, which results in very long inference times and high memory usage. Model compression technology such as unstructured pruning can prune a significant proportion of parameters without affecting accuracy, but efficient utilization of sparsity remains a challenge. Column combining compress unstructurally-pruned CNN models by combining multiple sparse columns in a convolutional filter matrix into a single dense column. In addition, pruning all but the largest magnitude weight in each row of the combined column further compresses the matrix effectively. However, previous work did not address the details of partitioning sparse columns to minimize the negative impact of additional pruning on the performance of the model. In this work, we first prove that the column partition problem is an NP-Complete problem. Next, we propose a column combining scheme based on simulated annealing and global unstructured pruning to minimize the adverse effects of additional pruning on model performance. We implement the acceleration of column-combined CNN models using the TVM AI compiler without special hardware support. The proposed scheme achieves more efficient model compression, leading to a 0.65% improvement in accuracy and a 1.24× faster inference time on VGG19 under 88% sparsity with the TinyImageNet dataset. © 2023 IEEE.

关键词： Simulated annealing

来源：评论

学校读者我要写书评

暂无评论

Effective Compression of Language Models by Combining Pruning and Knowledge Distillation

Effective Compression of Language Models by Combining Prunin...

引用

IEEE Annual International computer Software and Applications Conference (COMPSAC)

作者： Chi-Yu Chiu Ding-Yong Hong Pangfeng Liu Jan-Jan Wu Graduate Institute of Networking and Multimedia National Taiwan University Taipei Taiwan Institute of Information Science Academia Sinica Taipei Taiwan Department of Computer Science and Information Engineering Graduate Institute of Networking and Multimedia National Taiwan University Taipei Taiwan

ISBN: (数字)9798350376968

ISBN: (纸本)9798350376975

In recent years, Transformer has become an important architecture in language models and has achieved high performance in many natural language processing tasks. However, deploying Transformer architectures efficiently faces challenges because of their large model size and high inference time. Weight pruning is a prominent model compression technique that removes some weights in a model. However, after pruning, Transformer models require repeating the whole training process, including pre-training on a large generalized dataset and fine-tuning on a small downstream dataset, to recover the accuracy. The whole training process takes a long time and requires many computation resources. To address the challenge, this work proposes a pruning method that combines with knowledge distillation to avoid a long re-training time while recovering the accuracy. We use the N:m pruning method that is extended from NVIDIA's 2:4 pruning to compress the Transformer model. After pruning, we retrain the model by leveraging knowledge distillation to make the pruned model learn from the dense model. With this method, the pruned model can achieve comparable accuracy by using only downstream datasets and take much less time than traditional retraining. The experimental results show that DistilBERT in a 1:4 structure can achieve comparable accuracy on the SQuAD v1.1 and v2.0 datasets and a $\mathbf{1.7}\times$ speedups in inference time compared to the original dense model.

关键词： Training Weight measurement Adaptation models Accuracy Computational modeling computer architecture Syntactics

来源：评论

学校读者我要写书评

暂无评论

A Hybrid Approach for Binary Classification of Imbalanced Data

引用

International Journal of Computational Intelligence and Applications 2024年第3期23卷

作者： Tsai, Hsinhan Yang, Ta-Wei Wong, Wai-Man Kao, Han-Yi Chou, Cheng-Fu Department of Computer Science and Information Engineering National Taiwan University Taipei106319 Taiwan Graduate Institute of Networking and Multimedia National Taiwan University Taipei106319 Taiwan

Binary classification with an imbalanced dataset is challenging. Models tend to consider all samples as belonging to the majority class. Although existing solutions such as sampling methods, cost-sensitive methods, and ensemble learning methods improve the poor accuracy of the minority class, these methods are limited by overfitting or cost parameters that are difficult to decide. This paper proposes a hybrid approach with dimension reduction that consists of data block construction, dimensionality reduction, and ensemble learning with deep neural network classifiers. The performance is evaluated on eight imbalanced public datasets in terms of recall, G-mean, AUC, F-measure, and balanced accuracy. The results show that the proposed model outperforms state-of-the-art methods. © 2024 World Scientific Publishing Europe Ltd.

关键词： Classification (of information)

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：