Vijay V. Raghavan - US grants

A retrieval system model that leads to an integrated approach in which both local and global feedback naturally blend into a unified process is proposed. The novelty of this approach is that the relevance feedback of a particular query instance as well as the accumulated knolwdge from past queries are directly related to the performance of the present query. This model represents a cluster-based approach to information retrieval. The cluster-based approach is developed around a clustering technique that captures the users concept of closeness between documents. Since the clustering technique to support the retrieval process is developed. The principal advantage of the proposed model is that the retrieval performance can be directly influenced by the optimization criterion employed during clustering. The significance of our approach lies in the fact that document representation and the measure of similarity among doucments do not have to be prespecified in an ad hoc manner. Instead such design alternatives can be dictated by cluster-scope of application than just information retrieval.

0832420 University of Louisiana at Lafayette Vijay Raghavan

The University of Louisiana at Lafayette (UL Lafayette) seeks to partner with Oregon State University and form a new NSF Industry/University Cooperative Research Center (I/UCRC) for Visual Decision Informatics. The focus of this consortium will be to develop new visual and analytic methods that leverage modern computer hardware and software, and develop analysis and discovery tools that can be applied to this complex data-based decision-making process. The potential robustness of the approach utilizing highly immersive environments with flexible and real-time manipulation of large amounts of data has the potential to make breakthroughs in a variety of fields.

The proposed I/UCRC will provide a series of services complementary to other education, research and technology transfer units. It will provide students and faculty a platform to conduct industry-relevant research and gain valuable practical experience that otherwise cannot be gained from textbooks and research publications. The proposed center will recruit women and minorities through its ongoing outreach programs and through promotions with historically black colleges and universities in the region. The proposed center will also promote, catalyze and accelerate the commercialization of technology innovations.

I/UCRC for Visual and Decision Informatics (CVDI)

1160958 University of Louisiana at Lafayette; Raghavan
1160960 Drexel University; Hu

The Center for Visual Decision Informatics (CVDI) will develop new visual and analytic methods that leverage modern computer hardware and software, and develop analysis and discovery tools that can be applied to complex data-based decision-making process. The University of Louisiana at Lafayette (UL) and Drexel University (DU) are collaborating to establish the proposed center, with UL as the lead institution.

The proposed center will bring together analytic, visual, and perceptual techniques in order to support decision makers in industry and government by advancing the state-of-the-art in the research fields of Information, Visualization, Visual Analytics, and Automated Analysis. The proposed center's research agenda will be driven by the applications of information integration and decision support systems, 3D content creation and management in decision informatics, immersive visualizations, and human computer interaction technologies for training and education of workforce. To ensure that the research activities are transformative, software prototypes and frameworks developed through research will be tested in several application contexts by close collaboration with industry partners.

The proposed center activities will enhance the international competitiveness of the American industry in the problems and opportunities of leveraging data for evidence-based decision support. The center will help its potential members achieve higher productivity gains through the adopted data-driven decision making process, and will also impact the broader American industry through the publication and dissemination of the center's research results. Both universities are committed to including underrepresented groups and have active plans for addressing diversity. The center also intends to promote, catalyze, and accelerate the commercialization of technology innovations.

The proposed Visualization-Based Gap Analysis effort is aimed at providing an intuitive visualization and analysis techniques to provide analysts with the ability to understand what has happened within a domain, comprehend its current status and operations, and explore the impact of changes to the system. Link
Discovery is seeking to automatically predict what relations arise in the future between objects. The proposed research intends to reformulate the link discovery problem to include predicting the strength of the link as well as create techniques to detect and predict special categories of future links; such links that arise that bridge two dense clusters of objects. Link prediction for domains other than social media and specialized literature domains will be explored.

Predictive analysis to determine future events in an industry environment based on current and past data represents an outcome of tremendous economic and societal impact. Through establishment of more rigorous capabilities for Visual Gap Analysis and Link Discovery, industry may be better able to understand what has happened, comprehend what is the current state, and obtain some grasp of what is likely to happen, may be greatly enhanced increasing productivity and allowing focus on problem resolution. The proposed effort is supported by the Industry Advisory Board and will engage individual center members. Graduate students will be trained through the research and results will be integrated into courses and industry seminars and short courses.

This project, designing and developing an instrument that can support visual analytics on high volume, high velocity data streams, aims to offer an easy to use software interface for researchers to develop visual analytics applications that need a combination of stream processing, deep analytics, and visualization capabilities. The instrument provides the computational capacity and tight interconnection of systems to handle both real-time in-memory stream system processing and complex analytics, along with dedicated visualization processing. The instrument under development:
- Offers a customized computational system design expected to offer up to an order of magnitude performance over existing systems offered by commercial hardware vendors of business intelligent solutions.
- Advances knowledge and understanding in building highly-scalable big data platforms for decision makers or deals with big data generated from internet of things, and dynamic social networks that represent transportation routes, paths of disease outbreaks, social community networks, financial transactions, and recommendation graphs.
- Supports data mining and analytical needs of the Center for Visual and Decision Informatics (CVDI) and specific research projects to develop visual techniques on sensor data streams, and builds upon the experimental cloud infrastructure established in the Center for Advanced Computing Studies (CACS) Lab for InterNet Computing (LINC).

The real-time stream processing and analytics system comprises tightly interconnected processors with Remote Memory Access (RDMA) capabilities, low-latency Solid State Drive (SSD) with Infiniband Interconnect to support distributed in-memory data stream pre-processing, and analytics. The deep analytics nodes comprise data nodes that can handle efficient batch processing and the visualization processing node with high-end graphics cards to support visualization of massive data sets and advanced visualization venues. Supporting knowledge discovery and decision making requires a powerful computational infrastructure; without it, visual analytics on large complex dynamic graphs constructed from networks of sensors, mobile phones, social networks cannot be realized.

The instrument supports applications in many areas such as: disaster response, public safety, public health, cyber security, ecommerce and financial sectors. GENI (Global Environment for Network Innovations)-based connectivity to Lousiana's Optical Network Initiative (LONI) and Internet2 facilitates partnerships with researchers across other campuses nationwide and the US Ignite Community. The instrument will serve and be used by students for big data research and education. Utilizing the Louis Stokes-LA Alliance for Minority Participation (LS-LAMP), the project also facilitates participation of underrepresented students, thus increasing more participation in STEM areas. Furthermore, the instrument enables productivity for many researchers and educators.

This award will help the University of Louisiana at Lafayette transition its IUCRC Center for Visual and Decision Informatics (CVDI) into Phase 2 operation for the next 5 years. At the start of Phase 2, the CVDI IUCRC is expanding from three sites to 5 sites, with the addition of two new sites, and University of Louisiana at Lafayette will serve as the managing lead site of the Center for this phase. The Center seeks to continue its record of sustained accomplishments in terms of strengthening and growing industry partnerships, delivering innovation through intellectual property and publications, expanding research competencies through increased faculty participation, and broadening educational experience of students through participation in industry funded research.

The research program of the Center focuses on developing next-generation visual and decision support tools and techniques that enable decision makers to fundamentally improve the way an organization's information is interpreted and exploited. CVDI's area of research is industry-relevant, attracts membership and has the potential to improve US competitiveness in this area. The center intends to broaden participation through center engagement in university programs. Through participation in the Louis Stokes Louisiana Alliance for Minority Participation Program, of which the PI is the co-coordinator, this site will introduce undergraduate minority students to industry-driven research. CVDI will recruit students from neighboring Historically Black Colleges and Universities to join our graduate programs. The site will also participate in university outreach programs for K-12 students, such as the annual Engineering and Technology Expo Week and the Science Day.

This award provides travel support for 16 U.S.-based graduate students to participate in the 2017 International Conference on Data Mining (ICDM 2017), held in New Orleans, LA, November 18th - 21st, 2017 (http://www.icdm2017.bigke.org). The conference attracts new and original research, and some of the top data mining researchers from the U.S., and abroad to discuss their latest. The conference covers advancement in research in many topics relevant to data mining that include statistics, machine learning, pattern recognition, databases, data warehouses, data visualization, knowledge-based systems and high-performance computing. The proceedings of the ICDM conference will be distributed through the IEEE Computer Society and will be available through the IEEE Explore Digital Library.

Besides having a strong technical program, the conference features workshops in emerging topics on data mining, tutorials, panels, demos, and the PhD Forum. The PhD Forum is designed to provide an interactive environment in which PhD students can meet, exchange their ideas and experiences both with peers and with senior researchers from the data mining community. The ICDM organizers recognize the importance of recruiting, engaging and retaining students in data mining research given the importance of this topic in emerging field of data science. The conference participation enables the students to share their original research results with their peers and experts, learn about new algorithms and tools in the field of data mining, and obtain experience with application of data mining for various transdisciplinary problems. Participation of U.S.-based students is critical to developing U.S. competitiveness of future workforce in science and technology.

COVID-19 data, related to infection rates, at-risk populations, mobility, and commute dynamics are rapidly becoming available from several sources. However, there is a lack of interactive visual decision-making environments integrated with data-driven tools to help public health and community leaders understand how various factors such as physical distancing and other mitigation strategies, impact the spread of disease, help flatten the curve, enabling economic recovery while minimizing public health risk due to reopening. This project will develop visual analytic tools for tracking COVID-19 and propose balanced intervention strategies for effective containment of the outbreak.

The proposed visual analytics system integrates heterogeneous datasets and enables the application of relevant analytical models and data-engineering for decision support in a complex and evolving crisis. The objectives include the development of (1) forecasting models for recovery based on incidence, population vulnerabilities, mobility patterns, and mitigation activities, (2) social-media tools to understand public sentiment and risk perceptions, (3) visual interface for model-refinement & diagnosis through data engineering and visual analytics principles. The decision-making framework will offer new insights, close the gap between data and decisions, and is driven-by inputs from extensive partnerships & collaborations to improve reliability and usability.

The data-driven tools will help improve decision makersí understanding of disease dynamics from multiple variables. Epidemiologists could potentially leverage these insights to create higher-fidelity models based on interventional factors and their effect on population behaviors. Local authorities could also utilize the models to make life-saving decisions while minimizing impact to the economy. The project will enable new public and private partnerships including the City of New Orleans, and Industry Advisory Board of NSF Center for Visual and Decision Informatics. The project will benefit graduate and undergraduate students through hands-on research experience with the development of analytical products. The project outcomes will include analytics dashboards, source code, models, and data collected from multiple sources. The dashboards, project descriptions, and a list of data sources along with their metadata will be made publicly available on www.vastream.net for a period of two years. The public facing portion of the portal for COVID-19 component will be moved to Amazon cloud in event of disruptions from outages, for the duration of the project. A new public repository will be created on GitHub, and the source code and publicly available datasets will be made available on this project repository.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.