Neural mechanisms of visual motion detection. Nature Reviews 2008, Ann. Rev. Neurosci 2005. 

My area of expertise in neuroscience emphasizes computational systems. The idea is to think of the brain as a device that captures, stores and transforms information. Of course, that's what any computer does, but neural systems are a special kind, called massively parallel processors (MPP). MPP's exhibit emergent intelligence, which means their ability to compute isn't apparent in the details. So computational neuroscientists, by and large, try to understand the overall behavior of neural populations with mathematical models and electrophysiological techniques.

Structural computation in MT neurons: Nature 95, Nature 98.

My work has focused on neural circuits that process visual information, particularly motion cues. In one series of experiments, my collaborators and I studied the ability of primates and other vertebrates to sense 3D structure using visual motion cues. Humans' ability to do this has been recognized for nearly a century, but our studies were the first to pinpoint a brain area that produces structural information using visual motion signals. The area is called MT (middle temporal area) and it sits at the junction of the occipital, parietal and temporal lobes. It's about the size of a dime, and it's capable of processing information faster than any supercomputer ever built. If you wish to read more about these studies, please refer to Bradley et al, Nature 95 and Bradley et al, Nature 98.

Heading computation from optic flow. Bradley et al, Science 96.

Primates rely on visual motion cues to get a lot of information about their surroundings. Detecting moving objects is an obvious use, and as I discussed above, primates and probably other species can compute 3D structure from motion signals. Another, particularly interesting transformation is called direction of heading, or DOH. Even if everything around us is still, motion is created on the retina whenever we move around. As long as our eyes are not moving, our DOH and the pattern of optic flow are related by simple mathematics, and in the 1980's a Japanese group discovered that an area right next to MT, called MSTd, has properties consistent with these mathematics. But when our eyes are moving, and they usually are, the optic flow is a lot more complicated, and the mathematics are more complex. In that case the properties of MSTd seem incompatible with DOH computation. However we hypothesized that the problem could be solved by the same neurons as long as they effected a kind of coordinate shift. Coordinate shifts have been established in other neural computations but always in the spatial domain; in our case we postulated a shift in the spatio-temporal domain. Our experiments showed that this was almost certainly the case, suggesting that coordinate transformations in various dimensions could be a basic computational algorithm used in neural systems. Please refer to Bradley et al, Science 96 for details about the study.

Population coding in area MT. Purushothaman & Bradley, Nature Neuroscience 2005.

A large majority of neurophysiological studies are concerned with the way neural activities (action potential firing rates) reflect stimulus properties. But a fundamental question is, once these firing rates are produced, how does the brain interpret them? It has long been assumed that MT neurons are so sensitive, the brain could detect the direction of a moving stimulus by monitoring the output of a single MT neuron. But, with precisely controlled experiments, we showed that this could not be the case; instead, the brain appears to derive direction estimates from a pool of about 200 neurons, using a nonlinear version of the weighted average. This was an important result because it struck a blow from the popular "grandmother cell" theory, which implies that ultimately the things we perceive come down to the response of a single, highly specific neuron. For details see Purushothaman and Bradley, Nature Neuroscience, 2005.

Neuroscience Papers