检索结果-内蒙古大学图书馆

29th ACM International Conference on Information and Knowledge Management (CIKM)

作者： Vasudevan, Sriram Kenthapadi, Krishnaram LinkedIn Corp Sunnyvale CA 94085 USA Amazon AWS AI Seattle WA USA LinkedIn Sunnyvale CA USA

ISBN: (纸本)9781450368599

Many internet applications are powered by machine learned models, which are usually trained on labeled datasets obtained through user feedback signals or human judgments. Since societal biases may be present in the generation of such datasets, it is possible for the trained models to be biased, thereby resulting in potential discrimination and harms for disadvantaged groups. Motivated by the need to understand and address algorithmic bias in web-scale ML systems and the limitations of existing fairness toolkits, we present the LinkedIn Fairness Toolkit (LiFT), a framework for scalable computation of fairness metrics as part of large ML systems. We highlight the key requirements in deployed settings, and present the design of our fairness measurement system. We discuss the challenges encountered in incorporating fairness tools in practice and the lessons learned during deployment at LinkedIn. Finally, we provide open problems based on practical experience.

关键词： Fairness-aware machine learning scalable framework distributed computation LinkedIn Fairness Toolkit

来源：评论

学校读者我要写书评

暂无评论

Auto-Tiler: Variable-Dimension Autoencoder with Tiling for Compressing Intermediate Feature Space of Deep Neural Networks for Internet of Things

引用

SENSORS 2021年第3期21卷 896-896页

作者： Park, Jeongsoo Kim, Jungrae Ko, Jong Hwan Sungkyunkwan Univ Dept Elect & Elect Engn Suwon 16419 Gyeonggi Do South Korea Sungkyunkwan Univ Dept Semicond Syst Engn Suwon 16419 Gyeonggi Do South Korea

Due to limited resources of the Internet of Things (IoT) edge devices, deep neural network (DNN) inference requires collaboration with cloud server platforms, where DNN inference is partitioned and offloaded to high-performance servers to reduce end-to-end latency. As data-intensive intermediate feature space at the partitioned layer should be transmitted to the servers, efficient compression of the feature space is imperative for high-throughput inference. However, the feature space at deeper layers has different characteristics than natural images, limiting the compression performance by conventional preprocessing and encoding techniques. To tackle this limitation, we introduce a new method for compressing DNN intermediate feature space using a specialized autoencoder, called auto-tiler. The proposed auto-tiler is designed to include the tiling process and provide multiple input/output dimensions to support various partitioned layers and compression ratios. The results show that auto-tiler achieves 18% to 67% higher percent point accuracy compared to the existing methods at the same bitrate while reducing the process latency by 73% to 81%. The dimension variability of an auto-tiler also reduces the storage overhead by 62% with negligible accuracy loss.

关键词： collaborative intelligence deep feature compression inference partitioning autoencoder convolutional neural network deep learning machine learning distributed computation Internet of Things

来源：评论

学校读者我要写书评

暂无评论

Efficient 3-Party distributed ORAM 12th

Efficient 3-Party Distributed ORAM

引用

12th International Conference on Security and Cryptography for Networks

作者： Bunn, Paul Katz, Jonathan Kushilevitz, Eyal Ostrovsky, Rafail Stealth Software Technol Inc Los Angeles CA 90025 USA George Mason Univ Dept Comp Sci Fairfax VA 22030 USA Technion Comp Sci Dept Haifa Israel Univ Calif Los Angeles Dept Comp Sci Los Angeles CA USA Univ Calif Los Angeles Dept Math Los Angeles CA USA

ISBN: (纸本)9783030579890;9783030579906

distributed Oblivious RAM (DORAM) protocols-in which parties obliviously access a shared location in a shared array-are a fundamental component of secure-computation protocols in the RAM model. We show here an efficient, 3-party DORAM protocol with semi-honest security for a single corrupted party. To the best of our knowledge, ours is the first protocol for this setting that runs in constant rounds, requires sublinear communication and linear work, and makes only black-box use of cryptographic primitives. Our protocol also appears to be concretely more efficient than existing solutions. As a building block of independent interest, we construct a 3-server distributed point function (DPF) with security against two colluding servers that is arguably simpler and has better concrete efficiency than prior work. We also show how to distribute the key-generation protocol of this DPF (in a black-box manner).

关键词： Oblivious RAM (ORAM) distributed computation Function Secret Sharing (FSS) Secure Multiparty computation (MPC)

来源：评论

学校读者我要写书评

暂无评论

Merging Discrete Morse Vector Fields: A Case of Stubborn Geometric Parallelization

引用

ALGORITHMS 2021年第12期14卷 360页

作者： Lenseth, Douglas Goldfarb, Boris SUNY Albany Dept Math & Stat Albany NY 12222 USA

We address the basic question in discrete Morse theory of combining discrete gradient fields that are partially defined on subsets of the given complex. This is a well-posed question when the discrete gradient field V is generated using a fixed algorithm which has a local nature. One example is ProcessLowerStars, a widely used algorithm for computing persistent homology associated to a grey-scale image in 2D or 3D. While the algorithm for V may be inherently local, being computed within stars of vertices and so embarrassingly parallelizable, in practical use, it is natural to want to distribute the computation over patches Pi, apply the chosen algorithm to compute the fields Vi associated to each patch, and then assemble the ambient field V from these. Simply merging the fields from the patches, even when that makes sense, gives a wrong answer. We develop both very general merging procedures and leaner versions designed for specific, easy-to-arrange covering patterns.

关键词： discrete Morse theory gradient vector field distributed computation

来源：评论

学校读者我要写书评

暂无评论

distributed computational model for shared processing on Cyber-Physical System environments

引用

COMPUTER COMMUNICATIONS 2017年 111卷 68-83页

作者： Mora, Higinio Colom, Jose Francisco Gil, David Jimeno-Morenilla, Antonio Univ Alicante Specialized Processor Architecture Lab Dept Comp Sci Technol & Computat Alicante 03690 Spain Univ Alicante Dept Comp Sci Technol & Computat Lucentia Res Grp Alicante 03690 Spain Univ Alicante Dept Comp Sci Technol & Computat UniCAD Res Grp Alicante 03690 Spain

Cyber-Physical Systems typically consist of a combination of mobile devices, embedded systems and computers to monitor, sense, and actuate with the surrounding real world. These computing elements are usually wireless, interconnected to share data and interact with each other, with the server part and also with cloud computing services. In such a heterogeneous environment, new applications arise to meet ever-increasing needs and these are an important challenge to the processing capabilities of devices. For example, automatic driving systems, manufacturing environments, smart city management, etc. To meet the requirements of said application contexts, the system can create computing processes to distribute the workload over the network and/or a cloud computing server. Multiple options arise in relation to what network nodes should support the execution of the processes. This paper focuses on this problem by introducing a distributed computational model to dynamically share these tasks among the computing nodes and considering the inherent variability of the context in these environments. Our novel approach promotes the integration of the computing resources, with externally supplied cloud services, to fulfill modern application requirements. A prototype implementation for the proposed model has been built and an application example has been designed to validate the proposal in a real working environment. (C) 2017 Elsevier B.V. All rights reserved.

关键词： Cyber-physical systems Internet of things Mobile computing Modeling distributed computation

来源：评论

学校读者我要写书评

暂无评论

Building engineering safety risk assessment and early warning mechanism construction based on distributed machine learning algorithm

引用

SAFETY SCIENCE 2019年第10期120卷 764-771页

作者： Liu, Hongmei Tian, Guiliang Jiangsu Open Univ Sch Architecture Engn 832 Yingtian St Nanjing 210036 Jiangsu Peoples R China Hohai Univ Sch Business Nanjing 210098 Jiangsu Peoples R China

Cloud computing has become a hot topic in the industry and academic circles. Large-scale data processing is realized by centralizing system management of network resources, computing resources, and storage resources. Among them, the distributed computing platform greatly improves the productivity of programmers by abstracting the implementation details of distributed computing, and is widely used in building safety assessment. In this study, the construction safety evaluation index system was established for the construction site of common accidents, and the index weight was determined by the analytic hierarchy process. For the safety index evaluation task, an early warning mechanism based on distributed computing and extension theory to build cloud security management is established, which saves the overall calculation time and quantitatively evaluates the security status of the construction site. Finally, the method is verified by engineering examples, and the results show that the method is practical and effective. The method is convenient for computer programming, easy to operate and implement, and has strong applicability. Based on distributed computing and extended cloud model, not only can the overall state of construction safety be judged, but also potential security problems can be identified based on information feedback. The objective existence of hazards in the construction process can be grasped, which has important guiding significance for the construction process.

关键词： distributed computation Extension cloud theory Engineering Safety risk control Evaluation and early warning model

来源：评论

学校读者我要写书评

暂无评论

distributed Topology Manipulation to Control Epidemic Spreading Over Networks

引用

IEEE TRANSACTIONS ON SIGNAL PROCESSING 2019年第5期67卷 1163-1174页

作者： Xue, Dong Hirche, Sandra Tech Univ Munich Chair Informat Oriented Control D-80209 Munich Germany

This paper considers the problem of network structure manipulation in the absence of a central coordinator that collects network information and makes decisions. The primary focus is on the spectral radius minimization problem by removing/rewiring links to control epidemic spreading over networks. The resulting optimizations are generally combinatorial and NP-hard. The lack of the central base prevents us from solving such problems in a centralized fashion, thus driving the need for distributed computation and collaborative decision-making to support effective topology modification. In this paper, a distributed estimation scheme involving event-based communication and parallel algorithms is developed to enhance network capability against epidemics. Based on this low-complexity estimation algorithm, fully distributed strategies are proposed to enable individuals to sequentially discern the critical in-network contact, and to implement the desired link operation solely. We further extend the acquisitions to a more flexible framework based on a "coevolutionary networks" picture and discuss the optimality of solutions from an algebraic and topological perspective. Extensive simulation examples are presented to demonstrate the effectiveness of the proposed strategies.

关键词： Epidemic spread distributed computation link operation event-based control

来源：评论

学校读者我要写书评

暂无评论

Bounded Confidence Gossip Algorithms for Opinion Formation and Data Clustering

引用

IEEE TRANSACTIONS ON AUTOMATIC CONTROL 2019年第3期64卷 1150-1155页

作者： Linh Thi Hoai Nguyen Wada, Takayuki Masubuchi, Izumi Asai, Toru Fujisaki, Yasumasa Osaka Univ Grad Sch Informat Sci & Technol Osaka 5650871 Japan Kobe Univ Grad Sch Syst Informat Kobe Hyogo 6578501 Japan Nagoya Univ Grad Sch Engn Nagoya Aichi 4648603 Japan

This paper presents a bounded confidence gossip algorithm for describing the process of opinion formation over a communication network. Each agent in the network keeps a time-varying opinion vector (or state), which represents its opinion about a set of matters. A common confidence threshold is set for all of the agents. The states of agents in the network will be updated time by time according to an iterative procedure: At each time, 1) one agent is chosen randomly, then it chooses one of its neighbors on the communication graph to contact with;2) they exchange their states;and 3) if they have different states and the distance between their states is strictly smaller than the confidence threshold, they update their states as the average of the two. This algorithm converges almost surely to some equilibrium point such that any two adjacent agents either have the same state or have distinct states whose distance is no less than the confidence threshold. This is called the constant confidence threshold algorithm. An increasing confidence threshold algorithm, which repeats the constant confidence threshold algorithm several times with increasing confidence threshold, is also proposed. The algorithm is also convergent almost surely to some equilibrium point. Applicability of the method to clustering problems is shown through numerical examples.

关键词： Bounded confidence convergence analysis data clustering distributed computation gossip algorithm opinion dynamics

来源：评论

学校读者我要写书评

暂无评论

Spatial Matrix Based Clustering of Sparse Electric Power Networks

引用

SAIEE AFRICA RESEARCH JOURNAL 2019年第1期110卷 26-38页

作者： Rahman, Md Ashfaqur Venayagamoorthy, Ganesh Kumar Clemson Univ Holcombe Dept Elect & Comp Engn Real Time Power & Intelligent Syst Lab Clemson SC 29634 USA Texas A&M Univ Dept Elect Engn & Comp Sci Kingsville TX USA Univ Kwazulu Natal Sch Engn ZA-4041 Durban South Africa

distributed computation is an effective policy to increase the speed of the sparse networked systems. In a sparse network, clustering methods like k-means does not work directly as it cannot explore the connectivity of the system. To solve the problem, two modification methods are proposed in the existing graph and a new graph named Spatial Matrix is introduced in this paper. The proposed modification is a fast process and the computation time can be considered negligible compared to the rest of the process. Thus it preserves the ultimate objective of the distribution. It works as a pre-conditioning that can be used with a wide range of clustering and mathematical tools. With distributed state estimation of IEEE 14, 68, and 118-bus systems with automatic clustering, the effectiveness of the spatial matrix is demonstrated.

关键词： distributed computation network clustering sparse network spatial matrix WLS estimator

来源：评论

学校读者我要写书评

暂无评论

distributed Probabilistic ATC Assessment by Optimality Conditions Decomposition and LHS Considering Intermittent Wind Power Generation

引用

IEEE TRANSACTIONS ON SUSTAINABLE ENERGY 2019年第1期10卷 375-385页

作者： Avila, Nelson Fabian Chu, Chia-Chi Natl Tsing Hua Univ Dept Elect Engn Hsinchu 30013 Taiwan

This paper investigates a probabilistic assessment of available transfer capability (ATC) by optimality condition decomposition (OCD) techniques and latin hypercube sampling (LHS) method in power systems with penetration of wind energy resources. First, the ATC assessment is mathematically formulated as a non-linear optimal power flow problem. Then, an iterative decomposition-coordination methodology based on OCD techniques is conducted for distributed assessment of the ATC. In order to estimate the probability density function and empirical cumulative density function under wind power fluctuations, LHS method is utilized for obtaining samples of the integrated wind power sources. Next, wind power samples are appended into the proposed decomposition-coordination approach to provide a fast LHS-based Monte Carlo (MC) simulation of the ATC at the current system state. In order to provide a preliminary approximation of the range variation of the ATC before executing the MC technique, OCD-based distributed computations of the average-case, best-case, and worst-case scenarios subjected to wind power variations are developed by following three approaches: first, assuming that the average-case of the ATC occurs at the mean power output of integrated wind farms, second, adding box-constraints related to the power output of each wind farm in order to estimate the best-case scenario, and third, developing an iterative sensitivity-based scheme to estimate the worst-case scenario. Numerical experiments in the standard IEEE 118-bus system demonstrate the correctness of the proposed distributed probabilistic ATC assessment.

关键词： Available transfer capability (ATC) multiarea power systems optimal power flow decomposition-coordination distributed computation optimality condition decomposition (OCD)

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：