检索结果-内蒙古大学图书馆

Visual Explanation for Open-Domain Question Answering With BERT

ieee TRANSACTIONS ON visualization AND COMPUTER GRAPHICS 2024年第7期30卷 3779-3797页

作者： Shao, Zekai Sun, Shuran Zhao, Yuheng Wang, Siyuan Wei, Zhongyu Gui, Tao Turkay, Cagatay Chen, Siming Fudan Univ Shanghai 200437 Peoples R China Univ Warwick Coventry CV4 7AL England

Open-domain question answering (OpenQA) is an essential but challenging task in natural language processing that aims to answer questions in natural language formats on the basis of large-scale unstructured passages. Recent research has taken the performance of benchmark datasets to new heights, especially when these datasets are combined with techniques for machine reading comprehension based on Transformer models. However, as identified through our ongoing collaboration with domain experts and our review of literature, three key challenges limit their further improvement: (i) complex data with multiple long texts, (ii) complex model architecture with multiple modules, and (iii) semantically complex decision process. In this paper, we present VEQA, a visual analytics system that helps experts understand the decision reasons of OpenQA and provides insights into model improvement. The system summarizes the data flow within and between modules in the OpenQA model as the decision process takes place at the summary, instance and candidate levels. Specifically, it guides users through a summary visualization of dataset and module response to explore individual instances with a ranking visualization that incorporates context. Furthermore, VEQA supports fine-grained exploration of the decision flow within a single module through a comparative tree visualization. We demonstrate the effectiveness of VEQA in promoting interpretability and providing insights into model enhancement through a case study and expert evaluation.

关键词： Analytical models Transformers Task analysis data models Visual analytics Bit error rate Semantics Explainable machine learning open-domain question answering visual analytics

来源：评论

学校读者我要写书评

暂无评论

Toward Bi-directional In Situ visualization and analysis of Blood Flow Simulations With Dynamic Deforming Walls 12

Toward Bi-directional In Situ Visualization and Analysis of ...

引用

12th ieee symposium on large-scale data analysis and visualization (LDAV)

作者： Tishchenko, Nazariy Ferrier, Nicola Insley, Joseph A. Mateevitsi, Victor A. Papka, Michael E. Rizzi, Silvio Tan, Jifu Argonne Natl Lab Argonne IL 60439 USA Northern Illinois Univ De Kalb IL 60115 USA

ISBN: (数字)9781665491563

ISBN: (纸本)9781665491563

We present some preliminary work toward bi-directional in situ visualization and analysis of blood flow considering blood cell suspensions and dynamic moving walls to mimic the clot growth process. The work is implemented through integrating multiple open source packages including fluid flow solver, blood cells, and in situ visualization and analysis. It shows that in situ visualization enables us to examine and process the simulation results without writing data to files. In addition, preliminary results on bidirectional steering (still in progress) can accelerate engineers to explore the complicated stress environment experienced by clots of different sizes.

关键词： Deformable models Analytical models Simulation data visualization Bidirectional control Writing Aerodynamics

来源：评论

学校读者我要写书评

暂无评论

EmbeddingTree: Hierarchical Exploration of Entity Features in Embedding 16

EmbeddingTree: Hierarchical Exploration of Entity Features i...

引用

ieee 16th Pacific visualization symposium (ieee PacificVis)

作者： Zheng, Yan Wang, Junpeng Yeh, Chin-Chia Michael Fan, Yujie Chen, Huiyuan Wang, Liang Zhang, Wei Visa Res Palo Alto CA 94306 USA

ISBN: (纸本)9798350321241

Embedding learning transforms discrete data entities into continuous numerical representations, encoding features/properties of the entities. Despite the outstanding performance reported from different embedding learning algorithms, few efforts were devoted to structurally interpreting how features are encoded in the learned embedding space. This work proposes EmbeddingTree, a hierarchical embedding exploration algorithm that relates the semantics of entity features with the less-interpretable embedding vectors. An interactive visualization tool is also developed based on EmbeddingTree to explore high-dimensional embeddings. The tool helps users discover nuance features of data entities, perform feature denoising/injecting in embedding training, and generate embeddings for unseen entities. We demonstrate the efficacy of EmbeddingTree and our visualization tool through embeddings generated for industry-scale merchant data and the public 30Music listening/playlists dataset.

关键词： embedding analysis explainable AI machine learning visual analytics

来源：评论

学校读者我要写书评

暂无评论

Web-based visualization and Analytics of Petascale data: Equity as a Tide that Lifts All Boats 14

Web-based Visualization and Analytics of Petascale data: Equ...

引用

14th symposium on large data analysis and visualization

作者： Panta, Aashish Huang, Xuan McCurdy, Nina Ellsworth, David Gooch, Amy A. Scorzelli, Giorgio Torres, Hector Klein, Patrice Ovando-Montejo, Gustavo A. Pascucci, Valerio Univ Utah Salt Lake City UT 84112 USA NASA Ames Res Ctr Mountain View CA USA ViSOAR LLC Salt Lake City UT USA NASA Jet Prop Lab La Canada Flintridge CA USA CALTECH Pasadena CA 91125 USA

ISBN: (纸本)9798331516932;9798331516925

Scientists generate petabytes of data daily to help uncover environmental trends or behaviors that are hard to predict. For example, understanding climate simulations based on the long-term average of temperature, precipitation, and other environmental variables is essential to predicting and establishing root causes of future undesirable scenarios and assessing possible mitigation strategies. While supercomputer centers provide a powerful infrastructure for generating petabytes of simulation output, accessing and analyzing these datasets interactively remains challenging on multiple fronts. This paper presents an approach to managing, visualizing, and analyzing petabytes of data within a browser on equipment ranging from the top NASA supercomputer to commodity hardware like a laptop. Our novel data fabric abstraction layer allows user-friendly querying of scientific information while hiding the complexities of dealing with file systems or cloud services. We also optimize network utilization while streaming from petascale repositories through state-of-the-art progressive compression algorithms. Based on this abstraction, we provide customizable dashboards that can be accessed from any device with any internet connection, enabling interactive visual analysis of vast amounts of data to a wide range of users - from top scientists with access to leadership-class computing environments to undergraduate students of disadvantaged backgrounds from minority-serving institutions. We focus on NASA's use of petascale climate datasets as an example of particular societal impact and, therefore, a case where achieving equity in science participation is critical. We validate our approach by improving the ability of climate scientists to visually explore their data via two fully interactive dashboards. We further validate our approach by deploying the dashboards and simplified training materials in the classroom at a minority-serving institution. These dashboards, released in simpli

关键词： data visualization Petascale analytics data accessibility Equity in science Cloud computing Petabytes

来源：评论

学校读者我要写书评

暂无评论

KBMP: Kubernetes-Orchestrated IoT Online Battery Monitoring Platform

引用

ieee INTERNET OF THINGS JOURNAL 2024年第14期11卷 25358-25370页

作者： Chen, Qingyang He, Yinghui Yu, Guanding Xu, Chong Liu, Mingyang Li, Zhenming Zhejiang Univ Coll Informat & Elect Engn Hangzhou 310027 Zhejiang Peoples R China China Elect Power Res Inst Co Ltd Energy Storage & Novel Technol Elect Engn Dept Beijing 100192 Peoples R China

The rise in renewable energy has driven the widespread use of large-scale energy storage batteries, which makes the risk of overheating more threatening. To ensure battery safety, it is essential to build a monitoring system with a comprehensive evaluation of large quantities of batteries. However, existing battery management systems exhibit significant limitations in terms of monitoring scope, analytical precision, and transmission efficiency. As an applicable solution, cloud-edge technology is an advanced integrated method that provides low-latency data access, accurate analysis capabilities, and adjustable monitoring ranges. In this work, the Kubernetes-orchestrated battery monitoring platform (KBMP), which integrates Kubernetes and cloud-edge technology, is proposed to provide comprehensive battery management. Specifically, Kubernetes is used to ensure low latency in data transmission and analysis, while the K-Means clustering algorithm is applied to provide accurate thermal runaway (TR) warnings. To validate the performance of KBMP, four sets of real battery TR data are fed to test its accuracy and latency. The experimental findings reveal that KBMP is capable of providing battery TR warnings in advance within 30 min. Additionally, the platform concurrently decreases data transmission latency by up to 20% and reduces replica scaling latency by 50% compared to the platform without integrating Kubernetes.

关键词： Batteries Monitoring Cloud computing data visualization Servers data analysis data communication Battery monitoring platform cloud-edge technology K-Means clustering algorithm Kubernetes thermal runaway (TR) prediction

来源：评论

学校读者我要写书评

暂无评论

HSViz-II: Octet Layered Hierarchy Simplified visualizations for Distributed Firewall Policy analysis

引用

ieee ACCESS 2024年 12卷 936-948页

作者： Lee, Hyunjung Lee, Suryeon Kim, Kyounggon Kim, Huy Kang Korea Secur Comp Corp KOSCOM Seoul 07329 South Korea Seoul Womens Univ Dept CyberSecur Seoul 01797 South Korea Naif Arab Univ Secur Sci Dept Forens Sci Riyadh 14812 Saudi Arabia Korea Univ Sch Cybersecur Seoul 02841 South Korea

Enterprises typically install firewalls at communication points to their internal networks with the primary objective of protecting their core assets from external cyber attackers. This ensures unauthorized access is controlled and prevented. However, overly permissive policies and services with vulnerabilities can be exploited by attackers, providing them with pathways into internal systems. Therefore, firewall policies must be meticulously applied and managed. Given the significant ramifications of firewall policies, they must be continuously managed with high importance. As the number of policies increases and the amount of information to be processed grows, the process becomes complex, and there are limitations to managing policies from a human cognitive perspective. An increase in unmanaged misuse policies can inadvertently introduce security risks through unintended allowance policies. In the case of large-scale network networks operating multiple firewalls, a different form of misuse policy check and management is required compared to managing a single firewall policy. The proposed tool, HSViz-II, not only visualizes misuse of a single firewall policy but also visualizes four misuse cases in a distributed firewall environment, providing a detailed breakdown based on Octets. It displays the distribution of anomalous policies by dividing the Source IP into Octet Layers. For the four anomalous policy cases, it offers five views based on dividing the Source IP into Octet Layers and three overall views for upstream firewall, downstream firewall, and both, totaling 60 views. The processing speed for each function was measured using four sets of actual upstream and downstream firewall policies, comprising eight different firewall policies in total. Firewall operators can use this tool to grasp the distribution status of misuse policies in single and multiple firewalls and check the status of misuse policies by Octet. By offering a method for firewall operators to ac

关键词： Firewalls (computing) visualization data visualization Security Three-dimensional displays IP networks Distributed databases Guidelines Anomaly detection Firewall policy visualization policy analysis data visualization rule anomaly detection distributed firewall rule anomaly detection

来源：评论

学校读者我要写书评

暂无评论

Distributed Hierarchical Contour Trees 12

Distributed Hierarchical Contour Trees

引用

12th ieee symposium on large-scale data analysis and visualization (LDAV)

作者： Carr, Hamish A. Rubel, Oliver Weber, Gunther H. Univ Leeds Leeds W Yorkshire England Lawrence Berkeley Natl Lab Berkeley CA USA

ISBN: (数字)9781665491563

ISBN: (纸本)9781665491563

Contour trees are a significant tool for data analysis as they capture both local and global variation. However, their utility has been limited by scalability, in particular for distributed computation and storage. We report a distributed data structure for storing the contour tree of a data set distributed on a cluster, based on a fan-in hierarchy, and an algorithm for computing it based on the boundary tree that represents only the superarcs of a contour tree that involve contours that cross boundaries between blocks. This allows us to limit the communication cost for contour tree computation to the complexity of the block boundaries rather than of the entire data set.

关键词： data analysis Costs Scalability Distributed databases data visualization Clustering algorithms data structures

来源：评论

学校读者我要写书评

暂无评论

KD-Box: Line-segment-based KD-tree for Interactive Exploration of large-scale Time-Series data

引用

ieee TRANSACTIONS ON visualization AND COMPUTER GRAPHICS 2022年第1期28卷 890-900页

作者： Zhao, Yue Zhang, Jian Fu, Chi-Wing Xu, Mingliang Moritz, Dominik Wang, Yunhai Shandong Univ Qingdao Peoples R China Chinese Acad Sci CNIC Beijing Peoples R China Chinese Univ Hong Kong Hong Kong Peoples R China Zhengzhou Univ Zhengzhou Peoples R China Carnegie Mellon Univ Pittsburgh PA 15213 USA

Time-series data-usually presented in the form of lines-plays an important role in many domains such as finance, meteorology, health, and urban informatics. Yet, little has been done to support interactive exploration of large-scale time-series data, which requires a clutter-free visual representation with low-latency interactions. In this paper, we contribute a novel line-segment-based KD-tree method to enable interactive analysis of many time series. Our method enables not only fast queries over time series in selected regions of interest but also a line splatting method for efficient computation of the density field and selection of representative lines. Further, we develop KD-Box, an interactive system that provides rich interactions, e.g., timebox, attribute filtering, and coordinated multiple views. We demonstrate the effectiveness of KD-Box in supporting efficient line query and density field computation through a quantitative comparison and show its usefulness for interactive visual analysis on several real-world datasets.

关键词： data visualization Time series analysis visualization Rendering (computer graphics) Market research Clutter Three-dimensional displays Many time series density-based visualization interactive visualization for large-scale data

来源：评论

学校读者我要写书评

暂无评论

Effect of Weight Quantization on Learning Models by Typical Case analysis

Effect of Weight Quantization on Learning Models by Typical ...

引用

ieee International symposium on Information Theory (ISIT)

作者： Kashiwamura, Shuhei Sakata, Ayaka Imaizumi, Masaaki Univ Tokyo Tokyo Japan Inst Stat Math Tokyo Japan RIKEN AIP Tokyo Japan

ISBN: (纸本)9798350382853;9798350382846

This paper examines the quantization methods used in large-scale data analysis models and their hyperparameter choices. The recent surge in data analysis scale has significantly increased computational resource requirements. To address this, quantizing model weights has become a prevalent practice in data analysis applications such as deep learning. Quantization is particularly vital for deploying large models on devices with limited computational resources. However, the selection of quantization hyperparameters, like the number of bits and value range for weight quantization, remains an underexplored area. In this study, we employ the typical case analysis from statistical physics, specifically the replica method, to explore the impact of hyperparameters on the quantization of simple learning models. Our analysis yields three key findings: (i) an unstable hyperparameter phase, known as replica symmetry breaking, occurs with a small number of bits and a large quantization width;(ii) there is an optimal quantization width that minimizes error;and (iii) quantization delays the onset of overparameterization, which mitigate overfitting as indicated by the double descent phenomenon. We also discover that non-uniform quantization can enhance stability. Additionally, we develop an approximate message-passing algorithm to validate our theoretical results.

关键词： Contrastive Learning

来源：评论

学校读者我要写书评

暂无评论

Silent data Corruptions in Computing Systems: Early Predictions and large-scale Measurements 29

Silent Data Corruptions in Computing Systems: Early Predicti...

引用

ieee European Test symposium (ETS)

作者： Gizopoulos, Dimitris Papadimitriou, George Chatzopoulos, Odysseas Karystinos, Nikos Dixit, Harish D. Sankar, Sriram Univ Athens Athens Greece Meta Platforms Inc Menlo Pk CA USA

ISBN: (纸本)9798350349320;9798350349337

Silent data Corruptions (SDCs) due to defects in computing chips (CPUs, GPUs, AI accelerators) is a critical threat to the quality of large-scale computing in different application domains: cloud computing, high-performance computing, edge computing. Recent public reports by cloud hyperscalers have emphasized that apart from the usual suspects for SDCs (memory, storage, network), the heart of the computations, the processing elements of all types generate an unexpectedly large rate of SDCs which can cause erroneous calculations and severe information loss. We report, in a consolidated form, recent efforts to correlate early microarchitecture-level simulation-based predictions about the likelihood, rates, severity, and root causes of SDCs and large-scale in-field studies in cloud data centers. Early microarchitecture-level prediction of SDC characteristics (susceptible units, workloads, instructions) can shed light to the cryptic problem of SDCs. The findings of a diligent pre-silicon analysis can assist better understanding of SDCs and can thus drive effective protection decisions either at the hardware or at the software levels at deployment stages.

关键词： Silent data corruptions reliability cloud HPC edge microprocessors hardware reliability large-scale infrastructure microarchitectural simulation fault injection failure rates

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：