“One big challenge on the machine learning or AI side is really to create systems that are not just able to learn individual tasks, but instead have some semblance of common sense, pushing towards artificial general intelligence.”
Computational neuroscientist Joel Zylberberg is an assistant professor of physics and astronomy and researcher at York University’s Vision: Science to Applications (VISTA) program. His interdisciplinary work applies insights on how the human brain processes visual information to help program how a computer will execute the same set of tasks.
For example, he can take the brain activation patterns recorded when a volunteer completes activities and use those patterns to train an AI system. In particular, he hopes this approach will help create systems that — like people — can do more than one thing.
Through this greater breadth of experience and direct training from human brain patterns, the synthesis of concepts may allow machines to pick up aspects of common sense.
Based on these algorithms, Zylberberg is working on two main applications. The first is designing better bionic eye-type prosthetics to restore people’s vision when the light-sensing retinal layer of the eye is damaged or doesn’t function normally.
“The idea is being able to predict correctly what the outputs of the eye should be for any given visual input. We could then stimulate those same activity patterns and optic nerves, sending those signals into the brain in individuals with retinal dysfunctions.”
Being able to communicate in the same language as the brain could help make these connections in a way that feels natural for the user.
The second application is the interpretation of visual information when the environment changes very quickly, such as rapid changes in lighting. Where the human eye can readily adapt to these dynamic changes, it’s a major challenge for computer vision.
“We might be able to make better vision systems for applications, like autonomous vehicles, that need to work similar to our visual systems across a wide range of different luminance conditions,” says Zylberberg.
“So for example, when clouds start to cover the sky or when you go under a shadow, the light levels can change dramatically by factors of about 100 in fractions of a second, and that poses a real challenge for vision systems that need to work over those rapid, large changes.”
These big applications are rooted in big questions of human nature. Each insight that drives them forward also advances our understanding of ourselves.
“The passion that drives my research is really just a curiosity to understand the world around me and my place in it,” adds Zylberberg, “and that’s what keeps me happy.”
