Program in Neuroscience

Broadly speaking, I am interested in understanding basic principles underlying coding, computation and inference in the sensory nervous system. Sense organs measure physical signals and those signals are influenced by events in the environment. Animals are rarely interested in these sense data per se, but rather in the events that generate them. It is the job of the brain to infer those events-or at least some of their important aspects-from the set of raw measurements. However, sensory measurements are imperfect because of noise in the physical signal (e.g. photon shot noise), and because of bandwidth limitations of the receptors. Further, the relation between events and measured signals is often ambiguous. Therefore, the process of inference cannot be a simple fixed mapping or neural computation. Optimal inference requires that this computation depends on the joint statistics of sense data and events. The aim in my studies is to elucidate this process of inference at a fundamental quantitative level, both in an animal model system, and through analysis of the statistics of natural signals. These studies have both experimental and theoretical components. The electrophysiological experiments center on the visual system of the blowfly, where recordings are made from photoreceptors and from motion sensitive neurons. This system allows long term stable recordings, and the quality of the data makes it possible to put theories of neural coding and optimal processing to quantitative tests. One clear theoretical prediction is that the optimal processor is context dependent, and for this reason there is a substantial effort to understand neural processing in the animal's natural environment. Some of our electrophysiological experiments are therefore done outside. Using a specially developed setup we can record from a motion sensitive neuron while the fly is rotated along trajectories typical for free flying flies. Another series of experiment aims at sampling the joint statistics of camera input and camera motion in a real, natural environment, under natural conditions of locomotion. From a complete statistical desctiption of this re! lation we can derive an optimal motion sensor. Remarkably, this sensor shows specific biases in its estimation that are quantitatively similar to those measured in fly motion sensitive neurons under comparable conditions.