Today at Brown

Recognizing friends in a crowd or finding one’s car in a parking lot may seem simple, but visual object recognition is actually quite complicated. Researchers understand a great deal about how our eyes function, and they know that the brain’s visual cortex is responsible for processing visual data. Yet the neural mechanisms and codes by which our brains learn, categorize, and remember objects remain largely unknown.  

At the forefront of visual perception research, Brown Professor of Cognitive and Linguistic Sciences Michael Tarr was recently awarded a EUREKA grant from the National Institutes of Health (NIH) to study visual object recognition with a novel set of methods. EUREKA grants – the name is an acronym for Exceptional, Unconventional Research Enabling Knowledge Acceleration – are part of a new NIH effort to fund innovative, higher-risk biomedical research that may yield high-impact results.

With his $1.2 million, four-year EUREKA grant, Tarr will study visual cortical activity across millions of neurons in real time using functional magnetic resonance imaging (fMRI). He hopes to develop a better understanding of how our brains interpret visual data as meaningful objects – such as a classmate or a red Toyota.

Today at Brown spoke with Tarr about his research.

How did you become interested in visual perception research?

My interests drifted toward studying vision and artificial intelligence as I was finishing undergraduate work at Cornell in cognitive psychology. I have always been fascinated by artificial intelligence, particularly by [Carnegie Mellon University] Professor Herb Simon’s ideas on robotics and artificial vision, and by Raj Reddy, head of the Robotics Institute at Carnegie Mellon, who was my next-door neighbor during my teenage years in Pittsburgh.  

What are you looking for in your research, and why is it important?

I am interested in understanding how visual data is transmitted and processed between its point of detection – the eye – and its product – the recognition of visual data as discreet people or objects. One of the ultimate goals of the research is developing artificial vision systems. Another set of goals is clinical: individuals with dyslexia and autism have trouble recognizing visual patterns and expressions; research on facial expressions may lead to better physical therapy and compensation strategies for them. Finally, our work may lead to more interactive computer systems that could read a user’s facial expressions, recognize when they are frustrated, and offer appropriate tips.

FMRI images of human right visual cortex viewing birds and viewing faces show similar activity. (Credit: Michael Tarr) The EUREKA grant will enable you to pursue a new research direction. Please explain.

I feel that current research methods will not take our field anywhere new. Current neurophysiological research studies specific neurological events with high resolution, but it can only study tiny parts of the brain at a time – perhaps 1,000 or so out of billions of neurons in the visual cortex – and it may miss how neurons may be working in sync across different portions of the brain. On the other hand, fMRI techniques can study large areas of the brain, but they don’t “see” individual neurons, only general regions of summed neural activity.

Our new approach will use advanced fMRI methods to systematically map parts of the visual cortex by exploring how many different visual stimuli are processed in real time. We will first identify broad regions of the visual cortex that respond to a given stimulus, then zoom in on the specific parts of objects that are driving those groups of neurons. Real-time fMRI will allow us to adapt the images we show our research participants, adjusting which visual features are shown and isolating those features that maintain a similar neural response to that seen for the entire object. Eventually, our goal is to map much of the visual cortex, identifying which specific features drive which specific collections of neurons.

What outcome do you hope for?

One way of thinking about my approach is in terms of the children’s game “20 Questions.” When the brain receives visual data, it uses a set of visual “questions” to home in on the specific characteristics of a given object or scene in about 100 milliseconds. When we see a car, we very rapidly identify it as a non-living thing, as a thing with wheels, as a vehicle, then as an automobile and, if we care about such things, as a sedan, convertible, or SUV.

I want to play a version of 20 Questions with the human brain – taking a particular bit of it and asking it a series of visual questions in order to get it to reveal what it is visually coding about the world around us.

How do you feel about engaging in so-called high-risk research in which a desired outcome is not guaranteed?

It’s what makes science fun for me. I like to ask the big questions. There is a difference between high-risk and “wacko” questions. High-risk questions are reasonable and based on sound theories, but are just completely untested, and I think it is important to pursue them. The NIH in particular has been trying to make resources available for high-risk research.