For centuries, philosophers have debated the ways in which human beings can perceive and interact with the environment they live in. Neuroscience addresses the same basic conundrums. We see a chair across a room. But how does the brain tell us this chair is a chair? How does the brain recognize objects in the visual world? Over the last few decades, neuroscientists have discovered which sites in the brain may be involved in recognition. But they do not know how recognition works.
The focus of the Sinha Laboratory at MIT is to uncover the computational principles that underlie the brain’s recognition abilities. Research in the Sinha laboratory uses a combination of experimental and computational modeling techniques to discover:
- What is the nature of the object representations in the brain?
- How can object representations be learned from visual experience?
Lab director Dr. Pawan Sinha, Ph.D., associate professor of vision and computational neuroscience at MIT, divides his time between Cambridge, Massachusetts, and New Delhi. India is central to Sinha’s neurological work with the congenitally blind — the country is home to 30 percent of the world’s blind population, and 50 percent of these cases are either treatable or preventable.
In 2003, Sinha launched Project Prakash, a dual humanitarian and scientific endeavor, that provides medical intervention for children born blind and explores the nature of how object recognition builds when sight is first introduced to the brain. Sinha recently spoke to Brain World about how the brain learns to see, and how we learn most about vision from those not born with this faculty.
Brain World: Could you briefly explain how the brain learns to see?
Pawan Sinha: That’s like saying, “Summarize Proust!” I can describe some of the basic principles that people think that the visual system subscribes to. There is a very influential idea in the domain of visual neuroscience that essentially says that information from the eyes is not processed as a monolithic whole, but rather it’s split up into different kinds of attributes. There’s color, there’s motion, luminance, high-resolution information, low-resolution information.
And the belief is that these different attributes are being processed by different groups of neurons. The outputs of these neurons are eventually combined by some process that still remains mysterious. We don’t really know how that combination comes about.
BW: You mentioned splitting up visual attributes. How does that happen?
PS: If you record from neurons in different parts of the brain, you find different kinds of selectivities. Some neurons respond to moving stimuli, some respond more to color stimuli, and so on. It makes sense to say there is some splitting up of the signal. But exactly how everything is put back together — that remains unclear.
Even more intriguing is the question of how some neurons in the visual cortex achieve their exquisite selectivities. There are some neurons in the inferior temporal cortex which respond only to images of faces and respond very poorly to other objects. So it seems that even at a single neuronal level, you’re already getting evidence of recognition. It’s like that neuron is signaling, “Yes, the eyes are seeing a face,” versus they’re not seeing a face.
BW: Before you started working with blind subjects, your lab was interested in vision in other ways, wasn’t it?
PS: Absolutely. My research program is on high-level vision: how we recognize objects, and, more specifically, how we learn to recognize objects. How do we learn to make sense of the visual input from the eye? But it was a challenge how one might study visual learning from first principles. The only option, before Project Prakash came into being, seemed to be to work with newborn infants, because there is a creature learning to make sense of the visual world from very little proficiency, beginning proficiency.
But infants sleep for 18 hours a day, as I learned. And when they are awake, they are interested in other things, not your study. So baby work can yield some insight, but in order to really have a fuller program of research, one really needs to think about other options.