Data Science Ensemble Past and Present Research

The OmniSearch project is supported by an active NIH/NCI grant (U01CA180982). It aims to develo a semantic search tool to assist cancer biologists in unraveling critical roles of microRNAs (miRs) in human cancers in an automated and highly efficient manner. The project will handle the significant challenge of data sharing, date integration, and effective search in miR research in oncology. 

USA SoC Students:

Current:
Vikash Jha, Mohan Kasukurthi, Harrison J. Strachan

Outside Students:
Nisansa de Silva (University of Oregon)

USA Collaborating Partners:

Biology: Glen M Borchert
School of Computing: Jingshan Huang
Mitchell Cancer Institute: Zixing Liu, Ming Tan

Outside Collaborating Partners:

Judith A. Blake (Jackson Laboratory)
Dejing Dou (University of Oregon)
Karen Eilbeck (University of Utah School of Medicine)
Darren Natale (Georgetown University Medical Center)
Alan Ruttenberg (University of Buffalo - SUNY)

Itemset Tree is a data structure that, along with associated search algorithms, permits the ability to conduct targeted association mining. Association mining is a type of data mining that seeks to find correlations between multiple variables within a database. Current research efforts include improving the efficiency of targeted association mining and modifying the Itemset tree and algorithms to support advanced association and pattern mining.

USA SoC Students:

Current:
Lowell Crook

Graduated:
Vishal Bohara, Jay Lewis

USA Collaborating Partners:

School of Computing: David Bourrie, Tom Johnsten

Outside Collaborating Partners:

Alaaeldin M. Hafez (King Saud University)
Jennifer Lavernge (UL Lafayette)
Vijay Raghavan (UL Lafayette)

Social media has become a much discussed source of information; however, much of the analysis tends to be along the line of trending topics and terms. There is a growing emphasis on extracting addition types of information from the media; we have been pursuing two different efforts. One deals with detecting emerging events such as bomb threats, fires, road accidents, and drug recalls; the goal is to detect these events (and track them) within one to three minutes of their initial mention. The second centers around detecting new adverse drug reactions by analyzing Twitter data. This requires temporal reasoning, graph analysis and the ability to filter out spurious drug and reaction relationships.

USA SoC Students:

Current:
Murali Pusala (UL Lafayette)

Graduated:

Harika Karnati (UL Lafayette)
Satya Katragadda (UL Lafayette)

USA Collaborating Partners:

School of Computing: Ryan Benton

 

Collaborating Partners:

Chaomei Chen (Drexel University)
Weimao Ke (Drexel University)
Vijay Raghavan (UL Lafayette)
Xiaohua Tony Hu (Drexel University)

Action rules are constructs that provide guidance on what actions (i.e. changes to attribute values) should be made to convert a set of objects from an undesirable state to a more desirable state. For example, assume that you are seeking to determine what can be done to reduce the severity of traffic accidents. A potential action rule would state, if you add streetlights to a street with none, a significant number of accidents that result in severe injury would be reduced to accidents classified as minor. Current research includes the development of more efficient and effective algorithms for discovering action rules.

USA SoC Students:

Current:
Grant Daly, Shawyn Kane

USA Collaborating Partners:

School of Computing: Ryan Benton, Tom Johnsten

Contrast mining methods are designed to analyze data to discover patterns that occur frequently among one set of data objects, but relatively infrequently among other sets of data objects. These methods have been successfully used to analyze data for and in a wide variety of applications including change detection, object classification, and subgroup discovery. Designing efficient and effective contrast mining methods is challenging because of the time complexity of the problem. Current research includes the design and implementation of novel contrast mining methods for use in the context of high dimensional data and data streams.

USA SoC Students:

Current:
Glenn Santa Cruz

USA Collaborating Partners:

School of Computing: Ryan Benton, Tom Johnsten

Approximately 50 million people worldwide have epilepsy, making it one of the most common neurological diseases. To manage seizures, many patients require continuous use of medication. While helpful in managing seizures, the medication can alter the patients state of mind and reduce their quality of life.  We are investigating data-driven, theorem-based algorithms to accurately detect and predict seizures, with the aim to provide foundational tools for ambulatory treatment and assessment of patients who are affected by this ailment.  Our research uses novel feature analysis based on nonlinear time-delay embedding and is currently a semi-finalist project in INNOCENTIVE’s SUDEP Challenge.

 

USA SoC Students

Current:

Patrick Luckett

 Graduated:

William Ashbee

 

USA Collaborating Partners:

School of Computing:  J. Todd McDonald, Ryan Benton, Tom Johnsten

USA Epilepsy Center: Juan Ochoa

College of Arts and Sciences: Elena Pavelescu

 

Outside Collaborating Partners:

Dr. Lee Hively

Epitel

Epilepsy Foundation, through INNOCENTIVE’s SUDEP Challenge

 

In the realm of social robotics, JagBOT is designed to function as a mobile tour guide. With location and situational awareness, the intelligent system is able to interact in a variety of forms with human participants. This ability to use multiple sensors to create a signature for various points within an environment is a prime example of sensor fusion. Later work has begun to consider how JagBOT might be adapted to perform the functions of a flight attendant.

Current Student: 

 

Graduated:  

Clay Davidson

Clay Smith

John Licato

Michael Skinner (ENGR)

James Sakalaus (ENGR)

Hannah Becton (ENGR)

 

USA Collaborating Partners:

School of Computing: Dr. Michael Doran, Dr. W. Eugene Simmons

College of Engineering: Dr. Tom Thomas

 

Using a small scale mobile robot (LEGO NXT) a limited environment is created using RFID tags as waypoints. The robots are equipped with RFID readers, XBEE wireless communication and other sensors for navigation. Using simple learning algorithms the robots navigate the grid and are used to demonstrate cooperative and adversarial strategies.

Current:

George Clark

Graduated:

Alex Henderson

Jacob Maynard

Ed Baker

 

USA Collaborating Partners:

School of Computing Dr. Michael Doran

College of Engineering: Dr. Tom Thomas

 

 

Continuing the work on Intelligent Mobile Agent, the potential attacks on the learning algorithms are considered. As fully and semi autonomous robotic systems are created, the ability to adapt will be critical and rely on continuous unsupervised learning. This learning algorithm can be vulnerable to external attacks in a variety of ways: (i) hardware, (ii) firmware/OS and (iii) application levels. Attacks at any level would pose a serious danger as these system can and will impact all aspects of our lives. Work focuses on how these threats  might occur, be detected, and ultimately prevented.

Current:

George Clark

Graduated Students: 

 

USA Collaborating Partners:

School of Computing: Dr. Michael Doran, Dr. Todd Andel, Dr. Brad Glisson