The long-term goal of my research is to elucidate the set of interrelated computational problems that biological perceptual systems must solve in order to represent the natural sensory environment and extract useful information from it. A fundamental motivation of this work is to develop testable theories of perceptual function. The crux of this problem is that although we can observe features and properties of biological systems, the computational problems they solve are far from obvious. Work in my group spans many levels, from sensory coding to mid-level perception, in both vision and audition. We also have a deep interest in the mathematical algorithms and applications that stem from this work.

My research focuses on developing general algorithms for efficient signal processing and coding. Currently, I am working on spike-based algorithms for coding sounds, images, and video. This approach decomposes signals using kernel functions placed at particular points in space and time. The main challenges are the development of fast algorithms for signal decomposition and the learning of optimal kernel functions. This research promises a new class of coding algorithms that will dramatically improve coding efficiency. I am also working with Eizaburo Doi on mathematical analysis of robust coding in neural populations.

Currently, I study efficient encodings for acoustic data. I am implementing a model to capture non-linear structures in the spike codes produced by Smith and Lewicki. I also am interested in modeling neural adaptation mechanisms employed as perceptual data changes (e.g., in intensity or in other statistical properties).