Three faculty receive NSF CAREER awards

Assistant Professors Oliver Brock, Deepak Ganesan, and Erik Learned-Miller have each received five-year National Science Foundation (NSF) Faculty Early Career Development (CAREER) program awards, the NSF’s most prestigious award for new faculty members. Robot navigation, data gathered from smart sensors, and self-taught computer vision systems are the focus of their CAREER grants.

“Getting three CAREER awards in one year is outstanding, and indicates the quality of our junior faculty,” said Distinguished Professor and Department Chair Bruce Croft.

Brock received the CAREER award for his proposal “Motion Capabilities for Autonomous Mobile Manipulation.” His project explores the challenges involved in making robots independently mobile in various environments. Efforts in both industry and academia have led to quite advanced humanoid robots and mobile manipulators, said Brock, but there are still many factors that constrain these machines from fully operating in the everyday world. Brock will investigate generating robotic motion under these various constraints and will develop sensing techniques that provide feedback to robots about their surroundings.

At the low end of the frequency spectrum, he proposes an application of techniques from active learning and utility theory to global motion planning. These techniques permit the exploitation of structure inherent in any real-world motion planning problem to increase the efficiency of motion planning, said Brock. The resulting planning methods will satisfy the feedback requirements of autonomous mobile manipulation. “To obtain adequate feedback about the state of the environment, we will be developing novel sensing and modeling techniques to provide perceptual information,” said Brock. To address motion constraints that require high-frequency feedback, he plans to combine multi-objective control methods with planning approaches for constraint satisfaction. These planning methods overcome the susceptibility of control-based motion generation to local minima. Ultimately humanoid robots should be able to safely and reliably perform human-level tasks by themselves in unstructured environments.

Brock joined the Computer Science faculty in 2002 and co-directs both the Laboratory for Perceptual Robotics and the Bioinformatics Research Laboratory. He received his Ph.D. in Computer Science from Stanford University. His current research focuses on robotics, autonomous mobile manipulation, motion planning, and structural biology.

Ganesan’s proposal “Addressing Data and Energy Management Challenges in Hierarchical Sensor Networks,” garnered his CAREER award. In earthquake prone areas, sensors embedded in buildings to monitor vibration levels could predict if the building was becoming unsafe and inform people inside, but only if that information is stored properly, modeled accurately and easily accessible, said Ganesan.

His project tackles prediction techniques and storage systems for data from such sensors. It takes a fresh look at challenges in sensor networks in light of recent technology trends and experiences in pilot deployments. Technology trends indicate that the capacities of flash memories will continue to rise while their costs and energy consumption continue to plummet, said Ganesan. This will make it possible to equip sensor nodes with energy-efficient, high-capacity flash memory storage. In addition, pilot deployments have shown that scalable sensor network architectures will be hierarchical, and comprise hundreds of resource-constrained sensors but only tens of resource-rich sensor proxies. This motivates the need to develop methods to exploit resources at proxies while respecting constraints at sensors.

Ganesan’s research includes the design, prototyping and evaluation of archival storage subsystems for sensor nodes, algorithms to enable efficient access of large distributed archival sensor data, and compression techniques for efficiently retrieving such data. “This research will address systems issues as well as analytical underpinnings of a hierarchical sensor storage architecture and has strong ties to the fields of embedded systems, distributed systems and signal processing,” said Ganesan. Additionally, his project proposes an uncertainty-driven energy management architecture that unifies energy optimization across sensing, communication, routing, data processing and query processing tasks. The research uses prediction models and uncertainty as fundamental building blocks and builds a spectrum of energy-optimized services over this foundation.

Ganesan joined UMass Amherst in 2004 as an Assistant Professor and leads the Sensor Networks Research Group. He received his Ph.D. in Computer Science from the University of California, Los Angeles. Ganesan’s research interests include systems, networking and data management issues in sensor networks.

With his CAREER project, Learned-Miller aims to develop computer vision systems that are largely self-taught. Using modern learning techniques, it is now possible to teach computers visual concepts through example based learning. “But this process is time consuming and arduous,” said Learned-Miller. Often large data sets must be manually collected. Machines typically do not take advantage of previously learned knowledge when performing new tasks. And when confronted with a new situation, systems fail catastrophically. The goal of this research is to make it dramatically easier to teach vision systems new skills, and to design machines that can learn tasks faster by leveraging previously learned knowledge.

A central tenet of this work is that it is impractical to train vision systems one problem at a time, acquiring large training sets and developing training paradigms for each task to be learned. There are many scenarios in which training data are severely limited. And ideally, computer systems should be adaptive, and not have to be prepared for each new task, especially when these new tasks are similar to previous ones. Some specific areas of investigation include learning to recognize any particular car or face from a single example, simply by watching other cars or faces as they move about; developing software for robots to continuously explore the visual world and the interactions between vision and the other senses; and learning to recognize typewritten text in a font never seen before, without any training examples of that font. The common thread in these efforts is that they relieve the burden on the teacher of the computer. The final goal is to develop computers that can be taught simply and rapidly, and that can explore on their own.

Learned-Miller joined the Department in 2004. He received his Ph.D. in Electrical Engineering and Computer Science from the Massachusetts Institute of Technology. His research interests can be broadly categorized as applying ideas and methods from machine learning to problems in machine vision.

The CAREER program recognizes and supports the early career-development activities of those teacher-scholars who are most likely to become the academic leaders of the 21st century. Previous Department faculty CAREER award recipients include Micah Adler (2002), Emery Berger (2004), Mark Corner (2005), Brian Levine (2001), Sridhar Mahadevan (1995; awarded at Univ. of S. Florida), Kathryn McKinley (1996; now at UT-Austin), Prashant Shenoy (2003), Ramesh Sitaraman (1997), and Shlomo Zilberstein (1996).