The brain is probably the most fascinating physical system in the universe, but there is no established theory, even for its most basic aspects.
Stefano Fusi wants to design technology inspired by the human brain. As a step toward this goal, he is using math to better understand how the brain itself computes information, especially as related to problem solving, reasoning and decision-making. Because these brain-like machines would consume less electricity than today’s computers and be better at certain activities (such as recognizing images or driving cars), Dr. Fusi’s work could lead to technological advances that have thus far remained beyond our reach.Read more about Stefano Fusi, PhD >
If an engineer were to design the brain, she might keep things orderly, with one set of brain cells responding to only one kind of input while a second set responds to another. But it turns out that the brain is not that orderly. Instead, neurons, a type of brain cell, respond to all kind of things.
“Most engineers would consider all this noise and diversity a disturbance, something that you should get rid of,” says Stefano Fusi, PhD, a principal investigator at Columbia’s Mortimer B. Zuckerman Mind Brain Behavior Institute. “But actually the brain can take advantage of this, because it employs an entirely different computational paradigm.” Dr. Fusi, who is also a member of Columbia’s Center for Theoretical Neuroscience, attacks the problem of how exactly the brain computes.
A prime focus of his efforts is a feature of neurons called mixed selectivity. In addition to neurons that selectively pay attention to, say, the color of an object and not its shape, and others that selectively pay attention to shape and not color, we have a lot of neurons that pay attention to both. This lack of single-mindedness in the neuron is not a bug but a feature, and Dr. Fusi thinks it is a universal principle throughout the brain.
One way he has tested the idea is by looking at brain activity recorded from the frontal cortex, the area of the brain responsible for executive control, in animals while they perform a task. Using statistical analysis, he has found that when an animal responds correctly, populations of neurons show mixed selectivity and their responses are highly diverse, each responding to a different mixture of features. When the animal fails, the mixed selectivity component of the neural response is strongly reduced and significantly noisier. He has also used data that fellow Zuckerman Institute principal investigator Daniel Salzman, MD, PhD, has collected from the amygdala—a brain area central to emotion that is thought to be highly organized — and found a similar pattern.
This property is important in the brain because nearly every task you perform requires the consideration of multiple variables. Just reading this sentence requires combining one word with other words that came before, in order to extract meaning.
Dr. Fusi thinks he may be able to couple his mathematical tools with functional magnetic resonance imaging (fMRI) data on human brain activity to diagnose diseases for which there are currently no clear neural indicators. Some have suggested, for example, that autism is related to a lack of diversity in a specific type of neuron.
The work also has clear applications for developing what are called neuromorphic machines, computers that resemble the brain in both organization and functionality. The packing of more and more transistors onto a microchip, a trend described by Moore’s Law, will soon hit its physical limit. “One way to overcome this limitation is to take inspiration from the brain,” Dr. Fusi says, “especially if you consider the energy consumption of the brain.” A desktop computer cannot do with 500 watts what the brain can do with 20.
Dr. Fusi started out as a theoretical particle physicist, but, once he realized that the essential theories were already established, he lost interest. So he turned to the brain. “The brain is probably the most fascinating physical system in the universe,” he says, “but there is no established theory, even for its most basic aspects. People are still debating how information is represented in the brain.” Excited once again, he has a new mission: to break the neural code.