Global Resource Navigation:

Skip to Main Content

Apply to the University of Wyoming apply now

Global Resource Navigation

Visit Campus
Download UW Viewbook
Give to UW

Nature Article Highlights Clune’s Research

October 7, 2016
Jeff Clune (left) discusses his research with Wyoming Gov. Matt Mead.
Jeff Clune (left) discusses his research with Wyoming Gov. Matt Mead.

An article published Oct. 5 in Nature highlights the research that Computer Science Assistant Professor Jeff Clune and his team perform at the University of Wyoming.

Nature is an international weekly journal of science. The article is titled “Can We Open The Black Box of A.I.?” Clune refers to this portion of his work as “artificial intelligence neuroscience,” or trying to reverse engineer how machines, such as Deep-Neural Networks (aka Deep Learning), perform tasks.  This artificial intelligence powers everything from self-driving cars to websites that recommend products on the basis of a user's browsing history. The full article can be found here.

Davide Castelvecchi, the author of the article, writes: “Eventually, some researchers believe, computers equipped with deep learning may even display imagination and creativity.” The article goes on to detail the steps researchers are taking to understand networks.

The first artificial neural networks were created in the 1950s. But understanding how they operate is difficult. Like the human brain, memory is encoded in the strength of multiple connections, rather than stored at specific locations, like a conventional database.

As a result, “Even though we make these networks, we are no closer to understanding them than we are a human brain,” Clune says.

Clune's team discovered in 2014 that neural networks are easy to fool with images that to people look like random noise, or abstract geometric patterns. For instance, a network might see wiggly lines and classify them as a starfish, or mistake black-and-yellow stripes for a school bus. Interestingly, patterns trained to fool one neural network also tend to similarly fool all neural networks, even those trained on different data sets.

Researchers have proposed a number of approaches to try to solve the “fooling” problem, but so far none of them have solved it. Clune believes the issue could be dangerous in the real world, with scenarios including hackers sending a self-driving car veering into a billboard that it thinks is a road, or trick a retina scanner into giving an intruder access to the White House, thinking that the person is Barack Obama.

“We have to roll our sleeves up and do hard science, to make machine learning more robust and more intelligent,” Clune says.

Clune discussed the issues further in a Nature podcast, which can be found here.

Clune Earns Fellowship

College of Engineering and Applied Science Dean Michael Pishko announced Thursday that Clune was appointed as the Loy and Edith Harris Faculty Fellow in Computer Science for the next three years. That position comes with annual funding to advance his research efforts.


Share This Page:

1000 E. University Ave. Laramie, WY 82071
UW Operators (307) 766-1121 | Contact Us | Download Adobe Reader

Twitter Icon Youtube Icon Instagram Icon Facebook Icon

Accreditation | Emergency Preparedness | Employment at UW | Gainful Employment | Privacy Policy | Accessibility Accessibility information icon