Project Areas

Mobile and Sensor Networks

Buses and turtles have a lot in common, and it is not just their sometimes crawling pace. Brian Levine, Mark Corner, and Emery Berger are building sophisticated networked systems inside buses and on the shells of turtles. The bus system, known as DieselNet, provides a useful service for bus passengers and is the largest known test bed for research in disruption-tolerant networks (DTNs). DTNs allow for routing in networks that are unstable due to mobility, low node density, short radio range, intermittent power, environmental interference, and even malicious intent. These sorts of problems commonly occur when an infrastructure is destroyed by natural disaster or war, but they also occur in wildlife monitoring, sensor networks, and underwater acoustic networks. DieselNet, operational since May 2004, currently consists of 40 buses that travel around UMass Amherst and surrounding Hampshire County.

The Wood Turtle (Clemmys insculpta), found throughout the Northeast, lives along streams and in nearby woodlands. Its numbers are dwindling through loss of habitat and highway mortality, but conservation efforts to study the creatures in their natural habitat have been hindered by a lack of tracking data. Researchers currently track turtles manually using radio telemetry and are practically limited to recording a location every two to three days per animal. In order to more accurately understand how these turtles behave and use their habitat, Corner is developing a new tracking system to collect more frequent and detailed data. Engineering a sensor suitable for the back of a turtle highlights many of the challenges of a DTN device.

Privacy and Security of Implantable Medical Devices

Kevin Fu is advancing the understanding of security and privacy in pervasive computing. Implantable medical devices (IMDs) are an application where Fu is analyzing threats, testing power-efficient algorithms, and developing methods for usable security. Wireless embedded devices implanted within the body have revolutionized treatment of chronic disease with electrical therapy. While much research explores the safety of IMDs, there is limited knowledge and understanding of the security and privacy of devices such as pacemakers and implantable cardiac defibrillators (ICDs). This missing knowledge is important to discover because the latest IMDs provide advanced care by using the Internet and wireless connectivity. Ensuring security and privacy of an IMD is challenging because, like Radio Frequency Identification (RFID) and sensors, IMDs have only nomadic network connectivity and limited power and computation — making traditional cryptographic approaches infeasible. Unlike other devices, IMDs that fail prematurely require surgical intervention and expose patients to risk of infection, stroke, and death. Traditional approaches such as controlling access with a password or cryptographic key could delay crucial emergency patient care. Furthermore, exploring IMD security and privacy requires diverse skills from computer systems, embedded devices, and electrophysiology. Since changing a battery requires invasive surgery, a fundamental problem is understanding how to prevent unauthorized draining of the battery. Second, telemetry and therapy settings must remain private and integrity protected for safety and patient acceptance. This interdisciplinary project will introduce undergraduates to research methods for security and privacy of pervasive and invasive computation, and will involve collaboration with cardiologists from the Harvard Medical School.

Applying Computer Techniques to Social Networks: Analysis of the Social Network Among Professional Football Coaches

David Jensen applies data mining techniques to social networks to discover relationships. One interesting application is to professional football coaches. The interactions of professional football coaches and teams in the National Football League (NFL) form a complex social network. This network provides a great opportunity to analyze the influence that coaching mentors have on their protégés. Jensen uses this social network to identify notable coaches and characterize championship coaches. He also utilizes the coaching network to learn a model of which teams will make the playoffs in a given year. Developing comprehensive models of complex adaptive networks, such as the network of NFL coaches, poses a difficult challenge for researchers. Analysis of the NFL identifies three types of dependencies that any model of complex network data must be able to represent.
At the end of every National Football League (NFL) season, underperforming teams seek a coach with the combination of skills that will help the team win the most games in the future. That combination, however, is difficult. The primary alternative is examining a coach’s coaching “ancestry.” Considering this social aspect of coaching gives rise to coaching trees. Rather than considering coaching trees individually, Jensen examines the network of overlapping coaching trees to identify characteristics of successful coaches and the NFL as a whole. He also has demonstrated the effect that a coaching staff has on the success of a team. Future study should lead to deeper insight into the NFL as well as a collection of tools designed to more effectively model these types of data.

Computational Biology

The Computational Biology Lab, led by David Kulp and Oliver Brock is concerned with computational and modeling problems behind many of the major questions in molecular biology and genetics. Bioinformatics research involves the use of machine learning, modeling, and optimization techniques for DNA and protein sequence analysis. We maintain a mix of theoretical and applied experimental research with projects ranging from the technical development of novel microarrays to the theoretical intersection of computational geometry and protein folding.
As an example, Kulp is investigating gene regulatory networks. Recent studies have shown that QTL mapping using gene expression as a phenotype may be helpful in the discovery of upstream transcription factors in a gene regulatory network. In this project we intend to use genetic marker data as additional evidence for constructing gene regulatory network. Using this marker data and expression data, he intends to find for each gene, expression Quantitative Trait Loci (eQTLs) in the genome which regulate the expression of the corresponding gene.
Bayesian Networks have been successfully applied to this problem in yeast and others have shown how these probabilistic models can incorporate multiple types of information, such as common sequence motifs and common expression. Our strategy is to develop Bayes Nets that jointly model marker and expression data and apply these models to crosses of inbred mouse strains.
Oliver Brock’s work in autonomous mobile manipulation combines mechanisms, sensors, and algorithms that one day will enable robots to build habitats on Mars, service satellites in orbit or assist the elderly in their homes with daily chores. Surprisingly, some of the insights developed in the context of robotics are proving useful in the seemingly unrelated field of structural molecular biology. Brock, along with other colleagues in his field, have come to realize that proteins, the organic machines inside every living cell that perform a variety of important biological functions, can be viewed as tiny robots. Each of these proteins consists of a sequence of amino acids. The protein folds into a three-dimensional shape that is unique for each amino acid sequence. Based on this shape, the protein performs its function by interacting with other molecules inside the cell. Oliver’s research in molecular biology applies insights from robot motion planning to protein structure prediction. In spite of the vastly different scale of robots and proteins, both of these problems amount to a very similar computational challenge requiring algorithms that can efficiently build a representation of a very high-dimensional space.

Learning Technologies

Beverly Woolf and her colleagues use machine learning to model prior student behavior, to learn what is effective and to develop new and more effective pedagogical strategies. Optimization might be directed at reducing a student’s time to achieve mastery or advance through a curriculum. This research focuses on designing computer-aided tutors to improve their own knowledge about individual students and pedagogy. This improvement is needed for many reasons. Tutors are let loose in a constantly changing environment under conditions that cannot be predicted. It is not feasible to build teaching systems that emulate master teachers given the variety of student interests, abilities and learning styles. Nor is it feasible to build a tutor for every new population of students, e.g., those with low cognitive or spatial ability. Imagine an intelligent tutor that teaches fractions to 8-10 year old children. Such a tutor typically assumes that all incoming students have a fairly standard set of mathematics skills and will acquire new knowledge in a fairly standard way. These assumptions are clearly not valid for all 8-10 year olds, let alone those that are younger (e.g., a mathematics superstar learning algebra at a young age) or older (e.g., a thirty-year old college student studying remedial mathematics). Wayang Outpost, a tutor developed at UMass Amherst by Carole Beal and Woolf uses graphics and animation to motivate students. Machine learning enables this tutor to identify which methods are effective with which students and how to modify its heuristics.