检索结果-内蒙古大学图书馆

Anti-Collapse Loss for Deep Metric Learning

IEEE TRANSACTIONS ON MULTIMEDIA 2024年 26卷 11139-11150页

作者： Jiang, Xiruo Yao, Yazhou Dai, Xili Shen, Fumin Nie, Liqiang Shen, Heng-Tao Nanjing Univ Sci & Technol Sch Comp Sci & Engn Nanjing 210094 Peoples R China Hong Kong Univ Sci & Technol Guangzhou Guangzhou 511458 Peoples R China Univ Elect Sci & Technol China Sch Comp Sci & Engn Chengdu 610056 Peoples R China Harbin Inst Technol Shenzhen Sch Comp Sci & Technol Shenzhen 518000 Peoples R China

Deep metric learning (DML) aims to learn a discriminative high-dimensional embedding space for downstream tasks like classification, clustering, and retrieval. Prior literature predominantly focuses on pair-based and proxy-based methods to maximize inter-class discrepancy and minimize intra-class diversity. However, these methods tend to suffer from the collapse of the embedding space due to their over-reliance on label information. This leads to sub-optimal feature representation and inferior model performance. To maintain the structure of embedding space and avoid feature collapse, we propose a novel loss function called Anti-Collapse Loss. Specifically, our proposed loss primarily draws inspiration from the principle of Maximal coding rate Reduction. It promotes the sparseness of feature clusters in the embedding space to prevent collapse by maximizing the average coding rate of sample features or class proxies. Moreover, we integrate our proposed loss with pair-based and proxy-based methods, resulting in notable performance improvement. Comprehensive experiments on benchmark datasets demonstrate that our proposed method outperforms existing state-of-the-art methods. Extensive ablation studies verify the effectiveness of our method in preventing embedding space collapse and promoting generalization performance.

关键词： Encoding Measurement Training Task analysis Computational modeling Image retrieval Image coding Deep metric learning image retrieval embedding space coding rate

来源：评论

学校读者我要写书评

暂无评论

Deadline-Aware Coded Computation Across Homogeneous Workers

引用

IEEE SIGNAL PROCESSING LETTERS 2023年 30卷 982-986页

作者： Mehrabi, Mehrad Ardakani, Maryam Haghighi Ardakani, Masoud Univ Alberta Dept Elect & Comp Engn Edmonton AB T6G 2R3 Canada

Distributed computing systems have been widely used in recent years to handle massive computations required by newly emerged machine learning algorithms and signal processing problems. In practice, a distributed computing system often receives multiple tasks each needs to be finished by a specific deadline. This necessitates use of a task scheduler which orders and prioritizes tasks executions. In this work, we consider task scheduling for a homogeneous distributed computing system with multiple matrix-vector multiplication jobs, and try to maximize the number of tasks completed before their deadlines. The main challenges in such a system are random task arrivals and random execution times due to the straggling effect. To address these challenges, we propose two task scheduling algorithms namely "simple greedy" and "farsighted greedy" and compare their performance with the ultimate upper bound, i.e., a genie-aided algorithm that knows the exact arrival and execution times of all tasks. Our simulation results demonstrate that the proposed algorithms can approach the performance of the genie-aided algorithm.

关键词： Coded distributed computing task scheduling random task arrival task allocation coding rate

来源：评论

学校读者我要写书评

暂无评论

On the Performance Advantage of Relaying Under the Finite Blocklength Regime

引用

IEEE COMMUNICATIONS LETTERS 2015年第5期19卷 779-782页

作者： Hu, Yulin Gross, James Schmeink, Anke Rhein Westfal TH Aachen UMIC Res Ctr D-52074 Aachen Germany KTH Royal Inst Technol Sch Elect Engn S-10044 Stockholm Sweden

We consider a two-phase relaying system with finite blocklengths. We study the performance difference of relaying under the finite blocklength regime as well as under the Shannon capacity regime. Most importantly, we are interested in the conditions that lead to a higher performance of relaying under the finite blocklength regime. We find that these situations are characterized by error probabilities of relaying, e.g., the overall error probability of relaying and the error probability of the bottleneck link of relaying. We identify scenarios where relaying outperforms direct transmission under the finite blocklength regime even if their performances are similar under the Shannon capacity regime. Moreover, we prove that under these scenarios the performance advantage of relaying is more significant with short blocklengths. Finally, numerical results are provided and discussed.

关键词： Finite blocklength relaying coding rate error probability

来源：评论

学校读者我要写书评

暂无评论

On the Impact of Redundant Subcarrier Energy Optimization in UW-OFDM

On the Impact of Redundant Subcarrier Energy Optimization in...

引用

4th International Conference on Signal Processing and Communication Systems (ICSPCS)/12th International Symposium on DSP and Communication Systems (DSPCS)/9th Workshop on the Internet, Telecommunications and Signal Processing (WTSP)

作者： Hofbauer, Christian Huemer, Mario Huber, Johannes B. Klagenfurt Univ Inst Networked & Embedded Syst Univ Str 65-67 A-9020 Klagenfurt Austria Univ Erlangen Nurnberg Inst Informat Transmiss D-91058 Erlangen Germany

ISBN: (纸本)9781424479078

Unique word - orthogonal frequency division multiplexing (UW-OFDM) is a novel OFDM concept that uses deterministic sequences, which we call unique words, as guard intervals instead of the conventional cyclic prefixes. Since unique words represent known sequences, they can advantageously be used for sychronization, channel estimation, but also for improving the bit error ratio (BER) behavior of a system. These UWs are created by appropriately loading so-called redundant subcarriers. In this paper, we investigate methods to improve the BER behavior of UW-OFDM by increasing the number of redundant subcarriers (while keeping the length of the UW constant). This gain in the BER performance comes with an increase of the redundancy part and thus with a decrease of the data rate. As such, we present methods to vary the coding rate in UW-OFDM. We present results for the additive white Gaussian noise (AWGN) channel as well as for indoor multipath environments to highlight the advantages of the proposed methods.

关键词： AWGN channel AWGN channels BER OFDM modulation UW-OFDM additive white Gaussian noise channel bit error ratio channel estimation coding rate deterministic sequences encoding error statistics indoor communication indoor multipath environments orthogonal fre

来源：评论

学校读者我要写书评

暂无评论

coding STrateGIES FOR CDMA PACKET DATA NETWORKS WITH REDUCED RANK MULTIUSER DETECTION, ARQ AND PACKET COMBINING.

CODING STRATEGIES FOR CDMA PACKET DATA NETWORKS WITH REDUCED...

引用

9th IEEE Workshop on Signal Processing Advances in Wireless Communications

作者： Mouhouche, Belkacem Artigue, Cedric Freescale Semicond F-31023 Toulouse 1 France

ISBN: (纸本)9781424420452

Reduced-rank Code Division Multiple Access (CDMA) multiuser detection has received a lot of attention recently. The output SINR provided by a reduced-rank receiver depends on the filter rank and is confined between the SINRs of the Single User Matched Filter (SUMF) and the Minimum Mean Squared Error (MMSE) receiver. In Packet CDMA data networks like HSDPA, the measure of performance is the throughput. Thus, previous results on reduced-rank MMSE receivers SINR cannot be used directly. In this paper, we study the throughput of CDMA Data networks using reduced-rank MMSE receivers, Automatic retransmission request (ARQ) and Packet Combining. We derive the expression of the system throughput as the function of the filter rank, the coding rate and the number of retransmissions. We present simulation results to highlight the behavior of the maximum throughput as a function of the filter rank and the coding rate. We then conclude on coding strategies to adopt depending on the signal to noise ratio and the receiver rank in order to maximize the system throughput.

关键词： Reduced Rank MMSE Filtering CDMA Packet Data Network coding rate

来源：评论

学校读者我要写书评

暂无评论

Signal De-multiplexing in Branch Metric Calculation for Spatially Multiplexed MIMO System

Signal De-multiplexing in Branch Metric Calculation for Spat...

引用

IEEE 75th Vehicular Technology Conference (VTC Spring)

作者： Sanada, Yukitoshi Keio Univ Dept Elect & Elect Engn Yokohama Kanagawa 2238522 Japan

ISBN: (纸本)9781467309905;9781467309882

This paper presents a signal de-multiplexing scheme in branch metric calculation of soft decision decoding for a spatially multiplexed MIMO system. In a conventional MIMO system, de-multiplexing is carried out over the symbols received by multiple antenna elements and is separated from decoding. Therefore, the de-multiplexing requires multiple uncorrelated antenna elements in order to realize a full rank channel matrix. This leads to the limitation of the system capacity by the number of the receive antenna elements. Instead of splitting de-multiplexing and decoding, in this paper, a de-multiplexing scheme at branch metric calculation in soft decision decoding is proposed. Based on ideal interleaving, independence among the coded symbols is assumed. Thus, the full rank channel matrix with the size of minimum free distance can be realized for signal de-multiplexing. As examples of the proposed system, the performance of repetition codes, block codes, and convolutional codes with MMSE de-multiplexing and soft decision Viterbi decoding on a Rayleigh fading channel is investigated. It is shown through numerical analysis that a lower BER can be achieved with a larger minimum free distance for the same normalized transmission rate that is given as the product of the coding rate and the number of multiplexed signal streams.

关键词： MIMO Channel Capacity coding rate Minimum Free Distance

来源：评论

学校读者我要写书评

暂无评论

Scaling Law Based D2D Wireless Ad-Hoc Networks in the Finite Blocklength Regime

Scaling Law Based D2D Wireless Ad-Hoc Networks in the Finite...

引用

IEEE Global Communications Conference (GLOBECOM)

作者： Zhang, Xi Zhu, Qixuan Poor, H. Vincent Texas A&M Univ Dept Elect & Comp Engn Networking & Informat Syst Lab College Stn TX 77843 USA Princeton Univ Dept Elect Engn Princeton NJ 08544 USA

ISBN: (纸本)9781728109626

A device-to-device (D2D) wireless ad hoc network architecture enables dynamic self-organizing communications among mobile users who can directly exchange information with their peers without a pre-determined network infrastructure. Moreover, finite blocklength coding (FBC) is the promising candidate technique to support time sensitive multimedia wireless networks services, where mobile users transmit short packets to upper-bound the transmission delay of video/audio traffic. The scaling law technique models the maximum D2D channel capacity as a function of the density of mobile users. Recent studies have integrated D2D wireless ad hoc networks with FBC theory to further improve the performance of 5G wireless ad hoc networks. However, how to model and analyze the capacity of D2D wireless ad hoc networks under the finite blocklength regime is not well understood and has not been thoroughly studied. To overcome these challenges, applying the scaling law technique, we derive upper-bounds on the coding rate of each D2D channel and the number of time slots needed to complete all D2D transmissions. Combining the D2D channel's coding rate with the number of time slots needed for all D2D transmissions, we derive the maximum aggregate throughput for wireless ad hoc networks with all mobile users using D2D communications while mitigating interference. We also develop a model where each D2D channel follows the Nakagami-m distribution, under which we derive the average aggregate throughput and its upper-bound. Finally, we evaluate our derived results in the D2D wireless ad hoc networks over finite blocklength regime through numerical analyses.

关键词： D2D communications scaling law ad-hoc wireless networks finite blocklength coding theory coding rate information density mutual information rate aggregate throughput

来源：评论

学校读者我要写书评

暂无评论

MultiBand-OFDM system with concatenated coding scheme

MultiBand-OFDM system with concatenated coding scheme

引用

65th IEEE Vehicular Technology Conference

作者： Sato, Hiroyuki Ohtsuki, Tomoaki Kashima, Tsuyoshi Jarot, Sigit Tokyo Univ Sci Grad Sch Sci & Technol Tokyo 162 Japan Keio Univ Dept Informat & Comp Sci Tokyo 108 Japan Nokia Japan Co LTD Tokyo Japan

ISBN: (纸本)9781424402656

Among new technologies, Ultra Wideband (UWB) is gathering interests as a short-reach transmission system. At present, as one of the prevailing UWB systems, the MultiBand - orthogonal frequency division multiplexing (MB-OFDM) system has attracted much attention where convolutional code (CC) is used. The main contribution of this paper is to evaluate the bit error rate (BER) and decoding complexity performances of MB-OFDM with concatenated coding scheme and Dual Carrier Modulation (DCM) on IEEE 802.15.3a UWB multipath channel model. We employ the CC for the inner code and the Reed Solomon (RS) code for the outer code in the concatenated coding scheme. From the results of our computer simulation, we show good combination of coding rates in the concatenated coding scheme at the data rates of about 320, 400 480 Mbps. We also show that MB-OFDM with concatenated coding scheme can achieve the better BER than MB-OFDM with only CC. In addition, we show that the BER improvement of MB-OFDM with concatenated coding scheme over that with only CC becomes larger as the data rate becomes higher.

关键词： MB-OFDM convolutional code Reed Solomon code coding rate

来源：评论

学校读者我要写书评

暂无评论

Practical Self-repairing Codes for Distributed Storage

Practical Self-repairing Codes for Distributed Storage

引用

IEEE Asia Pacific Cloud Computing Congress

作者： Hou, Hanxu Li, Hui Peking Univ Shenzhen Grad Sch Shenzhen Key Lab Cloud Comp Technol & Applicat Shenzhen Peoples R China

ISBN: (纸本)9781467329040

In distributed storage systems, projective geometric self-repairing codes (PSRC) are able to provide a repair mechanism, which can repair a failed node with only two nodes. However, the redundancy of PSRC is extremely large, which makes PSRC unpractical for distributed storage systems. In this paper, we propose a new family of self-repairing codes called practical projective geometric self-repairing codes (PPSRC), which is a generalization of PSRC. PPSRC allow tradeoff between the coding rate and self-repairing capacity, while possess the basic property of PSRC. In addition, we provide an evolutionary repairing process for PPSRC, of which the repairing bandwidth is less than both the repairing process in [5] and that of the minimal storage repair codes (MSR) [3]. Furthermore, we present the general reconstruction process of PPSRC and the reconstruction process is optimal in terms of downloading bandwidth. Finally, we demonstrate PPSRC surpass the existing erasure coding strategies in terms of complexity, repairing nodes, repairing bandwidth and self-repairing capacity, which makes the PPSRC very attractive for distributed storage systems.

关键词： self-repairing codes distributed storage coding rate self-repairing capacity complexity

来源：评论

学校读者我要写书评

暂无评论

Performance Evaluation of OFDM-based 256-and 1024-QAM in Multipath Fading Propagation Conditions 9

Performance Evaluation of OFDM-based 256-and 1024-QAM in Mul...

引用

9th International Conference on Ubiquitous and Future Networks (ICUFN)

作者： Ota, Tomoki Nakamura, Mitsutoshi Otsuka, Hiroyuki Kogakuin Univ Grad Sch Engn 1-24-2 Shinjuku Ku Tokyo 1638677 Japan

ISBN: (纸本)9781509047499

The use of higher-order modulation is an important approach for increasing data rates within a limited bandwidth. In this paper, we demonstrate the performance of the orthogonal frequency division multiplexing (OFDM)-based 256- and 1024-QAM in multipath fading propagation conditions by link-level simulation, as well as under static propagation condition. The BER performance is investigated for two types of propagation models, i.e., "extended pedestrian A" and "extended vehicular A" models, specified by 3GPP as parameters of the coding rates. Using these results, we validate the application of OFDM-based 256- and 1024-QAM to mobile communication systems.

关键词： Mobile communication OFDM 256/1024-QAM EPA EVA coding rate Multipath fading

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：