College of Computing News

MURI Launches at Georgia Tech

A single ant can’t accomplish much, but an entire colony can gather food, build nests, and protect the queen. When powerless individuals can accomplish powerful goals together, it’s called collective emergent behavior, and it’s something researchers in physics, robotics, and computer science are trying to understand better.

Harnessing collective emergent behavior could lead to exciting new initiatives in computing. More than 40 researchers gathered at Georgia Tech’s CODA building on August 22 to discuss these possibilities at the kick-off for a $6.25 million Department of Defense’s Multidisciplinary University Research Initiatives (MURI) grant to study the phenomena over a five-year period.

ADVANCE Professor in Computing Dana Randall (pictured at right) leads a diverse team, including mechanical and chemical engineering, physics, and computational science professors from Georgia Tech and three other universities.

The team also includes Dunn Family Professor in the School of Physics Daniel Goldman, Arizona State computational science and engineering Professor Andrea Richa, Massachusetts Institute of Technology (MIT) chemical engineering Professor Michael Strano, MIT physics Associate Professor Jeremy England, and Northwestern mechanical engineering Professor Todd Murphey.

“This really is the most exceptional group I’ve ever collaborated with, and collectively we’re trying to do something even greater than we’d do individually—true to the theme of MURI,” said Randall, who is also a professor in the School of Computer Science and co-executive director of the Institute for Data Engineering and Science.

The Possibilities of Emergent Behavior

Emergent behavior is when microscopic changes can impact an entire system, like in an ant colony or robotic swarm. Although this has been observed in physics and computation, there has been no formal theoretical framework to explain how it functions until this work. The researchers will use basic algorithms on simple machines to perform complex tasks to predict and design emergent behaviors within computation.

[RELATED: Research and Collaboration Thrive @ GT Computing]

“This MURI brings together Georgia Tech’s main areas of strength and expertise,” said Executive Vice President for Research Chaouki Abdallah. “The partnership with other universities makes it really exciting. Hopefully we will solve some of these problems.”

The MURI’s Goals

Throughout the day, researchers explained their respective plans to tackle each of MURI’s three aims:

·      identifying and predicting emergent computation

·      actively evolving systems

·      determining optimal design and control

According to Richa, the first step is one of discovery.

“From a local distributed algorithm point of view, it’s very difficult,” said Richa. “For example, it’s easy to introduce phase transitions globally, but the challenge lies in handling multiple concurrent waves of transition in the system.”

To do this, researchers must discover which experimental and theoretical characteristics create emergent computation. Then they must determine how to model system capabilities, find language for describing what is possible, and learn how things can be designed to perform certain tasks.

Once the systems are understood, researchers work to use them to perform directed initiatives. These include efficient computation, finding equivalences between computation and physical properties, and employing fluctuation reduction as a design principle for cultivating active matter.

Goldman will explore this through his "smarticles,’’ a collective of simply computed robots.

“We’re discovering how to make a task-specific robot made of non-specific task incapable robots,” he said.

The third aim intends to predictably manipulate the behavior by finding optimization-based principles to design and control emergent computation systems.

This portion of the project is about discovering limitations and possibilities to achieve collective emergent behavior of an ensemble. This includes determining metrics for emergence and physical constraints that limit the system, and exploiting the system to solve prohibitively complicated computations.

“Part of the excitement is not knowing what might be next,” Randall said.