III: Medium: Collaborative Research: Human-Computer Graph Exploration and Tele-Discovery

III：媒介：协作研究：人机图探索与远程发现

• 批准号: 1563816
• 负责人: Duen Horng Chau
• 金额: $ 60万
• 依托单位: Georgia Tech Research Corporation
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 2016
• 资助国家: 美国
• 起止时间: 2016-08-01 至 2021-07-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1563816&HistoricalAwards=false
• 关键词: III; Medium; Collaborative; Research; Human

The amount of information available to individuals today is enormous and rapidly increasing. People are constantly making sense of the world: scientists learning the literature in an unfamiliar field; analysts spotting abnormal activities in computer networks; and patients understanding their symptoms. From a user's perspective, the main issue is not about storage, or computing power, or large scale data processing. It is more about how to best amplify his or her limited cognition power to make sense of a large data corpus via "natural" interactive exploration. This project will undertake the challenge of computer-human interactive exploration of information-rich billion-scale network datasets. These include online social networks (who is connected to whom), online auctions (who is buying what), and intelligence analysis of communication patterns and network traffic. It will blend computer-human interaction principles and decomposable visualizations with new scalable exploration techniques that are driven by information-theoretic measures. Specifically, it will design and develop a prototype system, in which users will gradually build up an understanding of billion-scale network datasets. This research could fundamentally change how people make sense of data in many domains like scientific literature, cybersecurity, and consumer decision making. The findings could increase education effectiveness, rate of scientific discovery, and enable more literate, knowledgeable, and intelligent citizens.This project will combine multiple novel ideas synergistically, organized into four inter-related research thrusts: (1) Adaptive Local Exploration using Minimum Description Length principles (MDL), KL divergence and Combinatorial Discrepancy. (2) Pattern Tele-Discovery & Global Summarization via algorithmic teleportation tools. These will include mechanisms for querying, discovering, linking, and visualizing multi-attributed time-evolving network patterns. (3) Scalable Data Models & Algorithms to support the interactivity demands of the previous thrusts. The proposed tools will address storage layouts via Egonet Edge Partitions and distributed sparse and persistent multidimensional sorted maps. (4) The researchers will continually conduct multi-stage evaluations in key domains, working with users throughout the entire development process. These will include iterative interface development via in-person user studies, virtual lab studies, and longitudinal field trials. For further information see the project web site at:http://poloclub.gatech.edu/human-computer-telediscovery/

今天，个人可获得的信息量是巨大的，而且还在迅速增加。人们不断地理解这个世界：科学家在一个不熟悉的领域学习文献;分析师发现计算机网络中的异常活动;病人了解他们的症状。从用户的角度来看，主要问题不是存储、计算能力或大规模数据处理。它更多的是关于如何最好地放大他或她有限的认知能力，通过“自然”的交互式探索来理解大型数据语料库。该项目将承担信息丰富的十亿级网络数据集的人机交互探索的挑战。这些包括在线社交网络（谁与谁相连）、在线拍卖（谁在购买什么）以及对通信模式和网络流量的智能分析。它将把人机交互原理和可分解的可视化与由信息理论措施驱动的新的可扩展探索技术相结合。具体来说，它将设计和开发一个原型系统，用户将逐步建立对十亿级网络数据集的理解。这项研究可能会从根本上改变人们在科学文献、网络安全和消费者决策等许多领域理解数据的方式。本项目将联合收割机的多个新的想法协同，组织成四个相互关联的研究重点：（1）使用最小描述长度原则（MDL），KL分歧和组合离散的自适应局部探索。(2)模式远程发现全球总结通过算法的隐形传态工具。这些将包括用于查询、发现、链接和可视化多属性时间演化网络模式的机制。(3)可扩展的数据模型算法，以支持以前的推力的交互性需求。拟议的工具将通过Egonet边缘分区和分布式稀疏和持久多维排序映射来解决存储布局问题。(4)研究人员将继续在关键领域进行多阶段评估，在整个开发过程中与用户合作。这些将包括通过亲自用户研究、虚拟实验室研究和纵向现场试验进行的迭代界面开发。欲了解更多信息，请访问项目网站：http://poloclub.gatech.edu/human-computer-telediscovery/

项目成果

NeuroCartography: Scalable Automatic Visual Summarization of Concepts in Deep Neural Networks

• DOI: 10.1109/tvcg.2021.3114858
• 发表时间: 2021-08
• 期刊: IEEE Transactions on Visualization and Computer Graphics
• 影响因子: 5.2
• 作者: Haekyu Park;Nilaksh Das;Rahul Duggal;Austin P. Wright;Omar Shaikh;Fred Hohman;Duen Horng Chau

NeuralDivergence: Exploring and Understanding Neural Networks by Comparing Activation Distributions

NeuralDivergence：通过比较激活分布探索和理解神经网络

• 发表时间: 2019
• 期刊: PacificVis 2019
• 作者: Park, Haekyu;Hohman, Fred;Chau, Duen Horng
• 通讯作者: Chau, Duen Horng

Bluff: Interactively Deciphering Adversarial Attacks on Deep Neural Networks

• DOI: 10.1109/vis47514.2020.00061
• 发表时间: 2020-09
• 期刊: 2020 IEEE Visualization Conference (VIS)
• 作者: Nilaksh Das;Haekyu Park;Zijie J. Wang;Fred Hohman;Robert Firstman;Emily Rogers;Duen Horng Chau