I am interested in theoretical and computational neuroscience:

    How does the brain work?
    How does the brain solve problems that engineers have found intractable?

My colleagues and I approach these questions through the combination of two disciplines: neuroscience and engineering. In our work, each discipline motivates and informs research in the other. The brain routinely solves many problems that are difficult or impossible for engineered systems, such as visual object recognition or robust control of action. We attempt to understand the brain’s solutions to these problems so that we can create more advanced engineered systems. Conversely, theories of computation and representation provide clues into the nature of neural representation and processing. For example, theories of efficient coding provide an explanation for neural representation in vision and audition. Because our work combines two fields of research, our work yields two distinct results: a better understanding of the brain and more advanced computational principles. The following outlines some specific areas of my own research.

Learning and Generalization:

A key property of a learning system is its ability to generalize to new examples. The ability of brains to generalize effectively allows them to behave intelligently in an ever-changing environment. What are the computational principles that produce representations that can robustly generalize?

Visual Representation:

In order to interpret the visual world, it is useful for visual systems to separate the invariant information from variant information. The objects of the world are an example of the invariant information while their position, size, and illumination determine the variances in the visual world. Invariant information is useful for determining what is out there, while variant information is useful for interacting with the world. Furthermore, these separate types of information may inform one another. How should invariant and variant information be separated and how should these representations interact?

Intermediate Level Vision:

Hierarchical representation is a key organizing principle of the cortical primate visual system. The representations and computations at the intermediate levels in this hierarchy have remained a mystery. Can theories of visual representation inform us about the coding we should find in intermediate cortical visual areas, specifically area V4?