11/17/2020 Laura Schmitt, Illinois CS
Topics include ways to enhance team-based learning experiences, improve network performance and functionality, accelerate graph computations on distributed computers, and create a robot that combines the control intelligence of humans with the physical strength and endurance of machines.
Written by Laura Schmitt, Illinois CS
In October, several Illinois CS faculty received new NSF research grants that provide a combined $2.25 million in funding to address a range of topics including ways to enhance team-based learning experiences, improve network performance and functionality, accelerate graph computations on distributed computers, and create a human-like robot that combines the control intelligence of humans with the physical strength and endurance of machines.
Here’s a summary of each of these newly funded projects.
Designing algorithmic tools to enhance team-based learning – NSF Award #2016908
Instructors who want to incorporate teamwork into their courses often struggle with grouping students into effective teams, especially in classes with large enrollments. Often, instructors allow the students to form their own teams, which may not result in a successful learning experience.
Brian Bailey (PI) is collaborating with fellow CS faculty Karrie Karahalios and Darko Marinov and Education professor Emma Mercier on a $750,000 project that is developing new methods and tools for grouping students into teams with effective compositions given the learning goals of the course. Drawing on theories of learning, team composition, and team building, the project will also develop new activities to help newly formed teams have successful learning experiences through improved psychological safety and team identity.
The technological advances Bailey and his faculty colleagues develop will benefit students by assigning them to teams where they can best use their individual strengths and learn to value the strengths of other members of the team. As a result, students will be able to engage and interact with teammates more effectively.
Applying machine learning to real-time network control - NSF Award #2008971
Rate control algorithms play a key role in optimizing data transmission and video streaming across the Internet. If these algorithms send data too slowly, then users will experience reduced video quality, but sending data too quickly causes congestion that impacts the sender and other users.
Brighten Godfrey (PI) received a three-year, $500,000 grant to develop new approaches to rate control based on an area of machine learning known as reinforcement learning, which he will apply to adaptive bit rate video, which is used in web-based video, and to transport layer congestion control, such as transmission control protocol (TCP).
Godfrey and his students, along with collaborators at the Hebrew University of Jerusalem, will explore what level of complexity of learning algorithm is required to achieve high performance. Second, they will develop a “scavenger” rate control protocol that helps ensure that important applications get the best performance, while other network traffic “scavenges” the remaining bandwidth when possible.
For example, if a person is working from home and having a video conference with a client, or even just a virtual get-together with family, he or she would want those applications to have the best performance while less important activities like an automated software update take a back seat. The team will also use deep reinforcement learning in a novel way – to automatically find weaknesses in protocols and improve their robustness in unexpected environments.
At the end of the three-year project, they envision better performance for deployed protocols, which can improve interactive conferencing, augmented and virtual reality, IoT devices, and edge computing. In August, the project received a Facebook Research Award.
Accelerating large-scale graph computations on distributed platforms - NSF Award #2028861
Graph computations are used in computational biology applications, road and network management, product recommendation, and path-planning problems in robotics. Josep Torrellas (PI) and co-PIs Chandra Chekuri, Sasa Misailovic, and Edgar Solomonik received a $250,000 grant to develop technology that accelerates large-scale graph computations on heterogeneous distributed computers.
Their work involves a new approach to solving graph computations that relies on approximation techniques, which allows the computation to be more parallel without compromising correctness. They’ll tackle the problem by focusing on the synergies between algorithms, numerics, compilers, and computer architecture.
In the algorithms area, they are investigating efficient parallel graph algorithms by leveraging approximation, continuous optimization techniques such as linear programming, and the use of sparsification methods.
In the numerics area, they are developing distributed-memory libraries of sparse-matrix computations for approximate graph algorithms. These libraries include techniques in graph algorithms, sparse linear solvers, and numerical optimization.
In the compiler area, they are developing novel techniques for approximate computation of graph applications, as well as automated verification approaches to guarantee their correctness. In the computer architecture area, they are increasing the speed of the resulting sparse-matrix computations with novel hardware.
The team is collaborating with IBM, a leading developer of high-end computer systems that run graph problems.
Creating a human-like robot for search and rescue — NSF Award #2024775
Kris Hauser (co-PI) is working with João Ramos (PI), a professor of mechanical science and engineering, on a $750,000 grant to develop technology that combines the control intelligence of humans with the physical strength and endurance of machines. Specifically, the team will create a human-like robot with manipulatable arms that moves on wheels and can perform physically demanding tasks like pushing or lifting heavy objects. The robot will be controlled remotely using teleoperation, meaning that the operator will use full-body haptics to communicate input and receive multisensory feedback.
During the three-year project, the researchers will implement a whole-body haptic device and humanoid robot system, study whole-body teleoperation strategies for Dynamic Mobile Manipulation (DMM)—physical tasks that require a combination of forceful manipulation and agile locomotion—and improve DMM safety by creating predictive control schemes that follow the operator’s commands while preventing input that would damage the robot.
Applications for this technology include disaster relief, firefighting, police work, logging, and mining.
Read more about this research.