Center for Digital International Government

UMD Logo

Home

Mission

Projects

Events

Links


Projects


D3: Deception, Deterrence and Disclosures

In April 2013, Nobel Laureate Tom Schelling asked us whether disclosure of behavioral rules about the terror group Lashkar-e-Taiba could tip them off that we were wise to their behavior. The D3 project asks the question: Can we use strategic disclosure of behavioral rules of a terror group to elicit better behavior from them? We develop a model that the adversary uses to carry out attacks in the presence of a set X of behavioral rules disclosed by the adversary and characterize the lethality of such attacks. We then show how a defender can identify a set Dmax of rules to disclose such that the expected response of the adversary is minimally lethal and maximally defendable.

Please visit the project website for more information!

Project team

  • Aaron Mannes, University of Maryland
  • Edoardo Serra, University of Maryland
  • V. S. Subrahmanian, University of Maryland



BAIT: Behavioral Analysis of Insider Threat

Consider an organization that seeks to find malicious insiders who are potentially exfiltrating sensitive information from the organization. The BAIT project tries to identify behaviors that separate out malicious insiders from benign ones. A major problem in real-world BAIT-like situations is that only dumb insiders use the same methods as past malicious insiders who have been caught. Our focus in BAIT is on intelligent insiders who may try new types of attacks. We developed a synthetic game on Amazon Mechanical Turk and did a study with 795 users some of whom were asked to compromise a computer system and tested the effectiveness of 7 algorithms most of which bootstrapped up to learn models from almost no training data. The best algorithm gave us about 60% recall and 30% precision in this very difficult challenge.

Please visit the project website for more information!

Project team

  • Amos Azaria, Bar-Ilan University
  • Sarit Kraus, Bar-Ilan University
  • Michael Ovelgonne, University of Maryland
  • Ariella Richardson, Jerusalem College of Technology
  • Edoardo Serra, University of Maryland
  • V. S. Subrahmanian, University of Maryland


Gorilla Behavioral Analysis

For more than 40 years, the Dian Fossey Gorilla Fund International (DFGFI) has undertaken international gorilla protection and research programs set in place by its namesake founder, zoologist Dian Fossey. This dedication to gorilla conservation has generated a huge amount of data on gorilla behavior, including data on poaching activity in gorilla territory to social interactions between gorilla groups. With the support of a generous grant from the Richard Lounsbery Foundation, CDIG researchers at the University of Maryland have collaborated with researchers at Zoo Atlanta and DFGFI to build a user-friendly web application to allow for fast and easy querying and filtering of this immense database. CDIG is also working with Zoo Atlanta and DFGFI in order to use social networks of gorillas to learn behavioral models of gorillas' social behaviors and predict certain specific behaviors.

Please visit the project website for more information and timely updates!

For a brief explanation of The Dian Fossey Gorilla Fund International's mission, visit their website here.

Project team

  • V.S. Subrahmanian, University of Marylandlink
  • Tara Stoinski, Zoo Atlanta & Dian Fossey Gorilla Fundlink
  • John P. Dickerson, University of Maryland & Carnegie Mellon Universitylink
  • Anshul Sawant, University of Maryland
  • Chanhyun Kang, University of Maryland


Anti-Poaching Engine

Starting with efforts to curb rhinoceros and elephant poaching, CDIG researchers are developing scalable, geographically sensitive algorithms to automatically fly a set of drones and coordinate them with a set of on-ground ranger patrols. The task is complicated by several factors. First, such coordinated ground-flight patrols must take into accont the poachers' known past behaviors. Second, the coordinated ground-flight patrols must take terrain information into account (elevation data, trafficability data). Third, the coordinated ground-flight patrols must take into account the movement patterns of the animals. Our APE platform develops a coordinated theory for this purpose and will be tested in cooperation with a game park in South Africa.

Please visit the project website for more information!

Project team

  • V.S. Subrahmanian, University of Marylandlink
  • Tom Snitch, University of Maryland
  • Edoardo Serra, University of Marylandlink
  • Noseong Park, University of Marylandlink



Predicting Systemic Banking Crisis

In partnership with collaborators at the International Monetary Fund, we examine patterns of connectedness based on cross-country financial linkages and relate them to the onset of systemic banking crises during 1978-2010. We rely on both data mining techniques as well as standard econometric models for crisis prediction. Our data, representing cross-border banking system assets and liabilities for a large number of countries, are mapped into a time series of global banking networks. Leveraging the strength of data mining algorithms developed previously in our lab, we are able to learn about the relationship between network-based indicators of connectedness (such as degree, strength, and clustering coefficients) and the onset of crises. Using our framework we are able to predict most of the systemic banking crises in the 2007-period onward 1, 2, or 3 years in advance. The data mining exercise is supplemented by a more traditional econometric analysis in which the probability of a crisis is linked to a set of standard macroeconomic factors and measures of financial connectedness. The model augmented with connectedness measures yields a 25 percent improvement in the accuracy of crisis prediction over the model that uses traditional macroeconomic variables alone. We conclude that financial connectedness has early-warning potential, especially for the most recent wave of crises in advanced economies.

Please visit the project website for more information!

Project team

  • V.S. Subrahmanian, University of Marylandlink
  • Camelia Minoiu, I.M.F.
  • Chanhyun Kang, University of Maryland
  • Anamaria Berea, George Mason University



Activity Detection in Video

Video surveillance is now ubiquitous and is widely used in airports, train stations, banks, and many other locations. We first show how stochastic automata with temporal constraints can be used to express activities occurring in video. We then develop algorithms to effectively identify all portions of a video in which a given activity occurs with a probability exceeding a user-specified threshold. We develop the tMAGIC index that monitors multiple activities occurring in video streams. We also developed PASS (Parallel Activity Search System), the first parallel algorithm to automatically identify activities in video. PASS can reliably process more than 500K observations per second.

Please visit the project website for more information!

Project team



Activity Detection for Cybersecurity

We leveraged our work on video activity detection to also detect activities in cyber-domains. We first show that stochastic temporal automata can be used to express attack graphs that are frequently used to describe cyber-attacks. We then develop algorithms to effectively identify all parts of a security transaction log in which a given activity occurs with a probability exceeding a user-specified threshold. We develop the tMAGIC index that monitors multiple activities occurring in such timed observation streams. We also developed PASS (Parallel Activity Search System), the first parallel algorithm to automatically identify activities in timed observation streams. PASS can reliably process more than 500K observations per second.

Project team

  • V.S. Subrahmanian, University of Marylandlink
  • Edoardo Serra, University of Maryland


WISE

Worldwide Information System for Educationwebsite

WISE, the Worldwide Information System for Education, tracks over 4700 variables related to a host of education outcomes for over 200 countries. WISE's goal is to understand the conditions governing a variety of contextual variables (e.g., health care, conflict, investment, education policy) that influence various educational desirable development outcomes (e.g., enrolment rates, dropout rates, female completion rates).

Using data collected by authoritative sources—such as the The World Bank and UNESCO—WISE uses sophisticated algorithms to automatically learn a class of rules called SOMA-rules. SOMA-rules explain the relationship between contextual variables and outcome variables and present them in an intelligent manner using a simple, easy to use query interface.

Please visit the project website for more information!

For a brief explanation of WISE on the world stage, see the AAAS press release here.

Project team

  • V.S. Subrahmanian, University of Marylandlink
  • Romain Murenzi, University of Maryland & AAASlink
  • John P. Dickerson, University of Maryland & Carnegie Mellon Universitylink
  • Damon Earp, University of Maryland


STAND

Surveillance, Tracking and Analysis of Nutrition and Diseases Worldwide

With the increasing need to mitigate the negative impact of the global food, and financial crises on children, there is compelling need for fresh data on ECD (in child nutrition, health, early childhood education) to be used in planning for the response and decision making. University of Maryland Professor V.S. Subrahmanian and World Bank economist Marito Garcia have jointly proposed a systematic procedure, accompanied by a supporting software architecture called STAND, to collect and monitor global ECD related data in near-time, analyze that data with human analysts and software tools, and provide policy makers with an experimental environment within which they can learn about and simulate options they need to determine what interventions would provide the highest impact in combating early childhood disease.

With respect to this project, Dr. Hamadoun Toure, Secretary General of the International Telecommunications Union said, "I am delighted by the advances in the research in this field. This will, with no doubt, ensure the improvement of quality of life of millions of people in the developing world where quality health care is not within reach of everybody. Through the use of affordable ICT we can make the world a better place."

Project team

  • V.S. Subrahmanian, University of Maryland -- link
  • Marito Garcia, The World Bank -- link

Publications

  • V.S. Subrahmanian and Marito Garcia. STAND: Surveillance, Tracking and Analysis of Nutrition and Diseases Worldwide. Proceedings of the 4th African Conference on Early Childhood Development, Dakar, Senegal, November 2009. (pdf download)


Country Profiles on Early Childhood Development: Sub-Saharan Africa

. . . for the Global Early Childhood Progress Report

To date, there has been almost no systematic quantitative effort to track the state of early childhood status in Africa, and the state of national and sub-national policies towards children's health, education, and welfare in Africa. In this report generated in cooperation with the RISE Institute, Save the Children (USA), and the Consultative Group on Early Childhood Development, we are pleased tp announce the release of a draft report containing quantitative information on the state of early childhood development in 37 countries in Africa. We are working on the development of a detailed, complete report, together with a web site and social networking environment with detailed analytics that will help early childhood policy makers worldwide better serve children.

Project team

  • Emily Vargas-Baron, RISE Institute -- link
  • V.S. Subrahmanian, University of Maryland -- link
  • John P. Dickerson, University of Maryland -- link
  • Regina Lauricella
  • Nitika Tolani-Brown
  • Lenisa Joseph
  • John Dougherty


Improving College Level Education in Nigeria

UMIACS researchers in the Center for Digital International Government worked with the World Bank, the University of Lagos, and the Federal College of Education - Technology, to devise a method to deliver high quality undergraduate programs to Nigerian universities, improve the basic ICT infrastructure in Nigeria, and help form partnerships for research and education between Nigerian faculty and students and university faculty and students elsewhere.

Project team

  • V.S. Subrahmanian, University of Maryland -- link
  • Fritz McCall, University of Maryland -- link

Organizations

  • Center for Digital International Government (CDIG)
  • University of Maryland Institute for Advanced Computer Studies (UMIACS)
  • The World Bank
  • University of Lagos
  • Federal College of Education - Technology


Center for Digital Government, University of Maryland Institute for Advanced Computer Studies, College Park, Maryland, USA