Research

Distributed Information Processing

Distributed architectures for signal processing have become, with the advent of sensor networks, an important area of research in the DSP community. Wireless sensor and actuator networks (WSANs) extend the traditional notion of a sensor network to a distributed network of sensors and actuators that produces a coordinated response to environmental stimuli without centralized processing and control. To accomplish these tasks efficiently, the entire processing architecture---the organization of measurement, processing, control, and actuation---needs to be distributed. Our research investigates the theoretical underpinnings of such distributed signal processing environments, using as our guide the processing strategy employed by the brain.

Information Theory for Neuroscience

The "Information Theory" developed by Shannon was intended to describe how information can be represented and communicated by digital messages. However, the communication system model has many similarities to neural systems, and almost from its inception neuroscientists have been applying information theoretic concepts to the study of neural systems. We continue this work cautiously; since neurons are not digital systems, there are many difficulties in applying classical information theory to the study of neural systems. Nevertheless, we believe that information theory has a lot to offer in computational neuroscience. Our most recent work involves point-process models of neural discharges, and may have profound implications in the design of neural prosthetics.

Neural Population Codes and Statistical Dependence

Information is processed in the nervous system by random sequences of electrical spikes, usually modeled as point processes. Recent advances in neurobiology have enabled recording these signals from large groups of neurons simultaneously. Traditional analysis techniques have revealed that these signals can be non-stationary and statistically dependent in both space and time. In our research we are attempting to answer some fundamental questions about population codes, the ways in which neurons cooperate to encode information:

In order to answer these questions, we must develop new statistical techniques to identify and quantify dependencies in large groups of random variables. These techniques must be robust in the absence of reliable models for neural interaction, and must be accurate given the limited amount of data that can be obtained from live subjects.

Links