Registration: www.conf.purdue.edu/salvendy

With the advent of the digital era, information has emerged as a co-equal building block of modern enterprises—alongside other components such as human agents, tangible materials and financial resources. Thus, aptly, the modern economy is often described as information economy. The growing importance of information as a resource has led to renewed interest in information engineering—a discipline that seeks to foster a structured engineering approach to the mining, storage, transportation, sharing, management and utilization of information. Recognizing the vast impact that information engineering is poised to have on society and industries of the future, IE and non-IE experts will give presentations on related visionary themes. The symposium should be of interest to researchers and professionals in any discipline that works with big data.

Speakers and topics will include:

Tuesday, September 29, 2015

8:00 - 8:20 a.m. Welcome

8:30 - 9:30 a.m. Big Data, Big Methods, Big Potential, Big Challenges

J. Marc Overhage, Chief Medical Informatics Officer, Cerner Corporation

The healthcare enterprise creates a large volume of data which holds considerable promise to inform how care is delivered in order to improve the quality, efficiency and safety.  The volume and variation of the data continue to increase particularly with the expansion of advanced imaging, genomic, metabolomic, environmental, health system data and patient generated health data.  Most of the uses of big data analysis in healthcare are high stakes: people’s lives and large amounts of money can be put at risk when conclusions are wrong and the chance that conclusions from big data analyses are wrong are high.  In order to fulfill the promise of big data in healthcare we need to develop new approaches and establish rigorous standards for studies that use observational or real world data.

9:40 - 10:40 a.m. Strategy Mining for Behavior Anticipation and Shaping

Alok Chaturvedi, Krannert School of Management, Purdue University

The Reference World Information and Simulation Environment (RWISE) is an ultrascale, data-driven virtual world whose primary objective is to decode behaviors and strategies of different entities by examining the interrelated effects of national dynamics, socio-economic and geopolitical situations, leader predispositions, and citizen expectations, goals, and desires for well-being.  RWISE can scale to hundreds of millions of heterogeneous, intelligent agents representing the virtual populations, organizations, institutions, and leaders of over 85 nations. RWISE maintains a tightly coupled, “near real time” connection between the external socio-economic, geopolitical environment and its RWISE virtual counterpart.  RWISE performs this critical role through the use of a semantic, big data technology, called SST, to create and sustain this coupling. SST is a very large, highly distributed data repository fed from three primary sources:  knowledge obtained from semantic data and text mining from the Internet and published documents and reports, structured data extracted from open source and subscribed databases, and feedback from RWISE simulation updates.  Three-dimensional tagging, consisting of source, point of view, and a time stamp, is employed to each atom of data in SST. RWISE simulations allow users to employ computational experimentation to leverage the knowledge base by creating and executing historical or futuristic simulations to mine and validate strategies of different sides in order to develop and test policies and courses of actions.  

10:50 - 11:50 a.m. Divide & Recombine (D&R) with Tessera: High Performance Computing for Data Analysis

William Cleveland, Department of Statistics, Purdue University

The widely used term "big data" carries with it a notion of computational performance for the analysis of big datasets. But for data analysis, computational performance depends very heavily, not just on size, but on the computational complexity of the analytic routines used in the analysis. Data small in size can be a big challenge, too. Furthermore, the hardware power available to the data analyst is an important factor. High performance computing for data analysis can be provided for wide ranges of dataset size, computational complexity, and hardware power by the D&R statistical approach, and the Tessera D&R software implementation that makes programming D&R easy.

12:00 - 1:00 p.m. Lunch

1:10 - 2:10 p.m. Data Analytics for Cyber Forensics

Marc Rogers, Purdue Polytechnic Institute, Purdue University

The talk will examine how data analytics can be used in cyber forensics to assist with the problem of data overload. Historically cyber forensics has focused on data acquisition. However, with the relatively cheap cost of storage, the amount of data present in some cases is overwhelming and the traditional cyber forensics tools have not adjusted to this growth. The discussion will look at how data analytics is being currently used and how cyber forensics can leverage the advances in analytics from such fields as business intelligence and social network analysis.

2:20 - 3:20 p.m. Examples of Algorithmic Data Pipelines for Social Networks and Large Simulation Analysis

David Gleich, Department of Computer Science, Purdue University

We will survey how large datasets of social network data are mined for information about the communities underlying each node and to optimize the performance of distributed systems. We will also see examples of how a big data pipeline worked to analyze a large collection of simulation data at Sandia National Labs.

3:20 - 4:00 p.m. Tea Break

4:00 - 5:00 p.m. The Internet in Attention Overdrive: Can We Find a Cure for Information Overload?

Bruno Ribeiro, Department of Computer Science, Purdue University

In 1969, Nobel laureate and Turin Award winner Herbert A. Simon warned us that the newly designed computer mainframes might inadvertently increase society’s information overload problem. Simon could not have predicted the fierce competition for our attention of today’s Internet apps, installed ubiquitously on our personal computers, tablets, phones, and even watches. In my talk, I will show how the dynamics of online social media combined with a phenomenon known as “the network effect” create an app ecosystem that helps explain our current attention-depleted world. I will also discuss the implications of these findings on the life cycle of Internet startups and propose possible solutions to the problem.

Wednesday, September 30, 2015

8:30 - 9:30 a.m. Data Science in the Financial Sector

Mark Bennett, Global Investment Bank and University of Chicago

We discuss Data Science and survey some of the important Data Science models utilized in the Financial Sector. In order to simulate extreme events which can occur in the various financial markets, the subpopulations can be jumps or crashes in the market. While applying a non-Gaussian distribution is common practice for introducing these jumps, it is also reasonable to use two or more single-variate Gaussian distributions and combine them into a mixture model. We apply it to simulations from the foreign exchange markets, especially for the large jump in the Swiss Franc occurring in January 2015. We then switch gears over to the world of Network Science where Gaussian Graphical Models can be used to visual covariance inherent in market portfolios. These are just a small sample of the tools in the emerging field of Financial Analytics.

9:40 - 10:40 a.m. From Anecdotes to Analytics - How Data is Changing Healthcare and Hospital Operations

Tze Chiam, Value Institute, Christiana Care Health Systems

According to Institute of Medicine in 2012, when compared to other industries, the American healthcare industry falls short on fundamentals such as quality, outcomes, cost and equity. Despite having the largest health expenditure per capita compared to other developed countries, US healthcare lags behind its peers based on measures such as infant mortality and hospital efficiency. In order to improve and optimize the healthcare system, the Institute for Healthcare Improvement (IHI) developed the IHI Triple Aim framework based on 3 dimensions: Improving patient experience of care, improving the health of populations and reducing per capita cost of healthcare. To accomplish these goals, data and quantitative measures need to be in place. Today’s talk will focus on the changing healthcare landscape, selected hospitals’ initiatives for improving patient experience of care and health through the use of data, analytics, Industrial Engineering techniques as well as implementation science in large hospital systems. Projects discussed will include Geographic Cohorting of Hospital Medicine patients (versions 1 and 2), which utilizes analytical skills such as Discrete-Event Simulation, Statistical Process Control, Queuing Theory, as well as managerial skills including change management and Lean Six Sigma. Other current and future opportunities in healthcare research and implementation science will also be discussed such as prediction of diseases and complications, healthcare big data, population health, among others.

10:50 - 11:50 a.m. Distributed Statistical Inference with Compressed Data

Lifeng Lai, Department of Electrical and Computer Engineering, Worcester Polytechnic Institute

In the classic statistical inference problems, all data is available at a centralized location. With the explosion of the size of data set, it is increasing common that data is stored in multiple terminals connected by links with a limited communication capacity. In this scenario, terminals have to exchange compressed data and perform statistical inference using the compressed data. In this talk, we will discuss our recent work that addressed the following two questions: 1) For a given communication rate budget, what are the best compression and inference algorithms and their corresponding performance?; and 2) Suppose we would like to achieve an inference performance that is the same as that of the centralized scenario, what is the minimum communication rate required?

12:00 - 1:00 p.m. Lunch

1:30 - 2:30 p.m. Deep Learning for Computer Vision

Qian Lin, Emerging Technologies Lab, Hewlett Packard Laboratories

Image sensor market is growing rapidly, driven by strong demand from automotive applications, security and surveillance systems, and medical imaging. While the majority of image sensors are on smart phones capturing photos and videos for sharing and memory-keeping, the fastest-growing segments of image sensors require real-time computer vision to interpret images and take action. With new advances in deep learning, the accuracies of many computer vision tasks are now exceeding those of human vision. This further fuels the wide adoption of using computer vision for monitoring and real-time decision making.

In this presentation, Dr. Lin will give an overview of deep learning. She will then review various computer vision challenges that are now solved by training CNN models. She will next discuss how computer vision computation can be integrated into a big data platform and parallelized to process large amounts of image and video data. Finally, she will show several industrial applications of computer vision.

2:40 - 3:40 p.m. A New Method to Construct Large Gene Regulatory Networks Using Genetical Genomics Data

Dabao Zhang, Department of Statistics, Purdue University

Constructing whole-genome gene regulatory networks using genetical genomics data is challenged by limited computer memory and intensive computation. We propose a two-stage penalized least squares method to study regulatory interactions among massive genes, building up reciprocal graphical models on the basis of simultaneous equation models. Fitting a single regression model for each gene at each stage, the method employs the L2 penalty at the first stage to obtain consistent estimation of surrogate variables, and the L1 penalty at the second stage to consistently select regulatory genes among massive candidates. The resultant estimates of the regulatory effects enjoy the oracle properties. Without fitting a full information model, the method is computationally fast and also allows for parallel implementation. We demonstrated the effectiveness of the method by conducting simulation studies, showing its improvements over other methods. Our method was also applied to construct a yeast gene regulatory network.

3:40 - 5:00 p.m. Reception