INFORMATION DISTORTION AND NEURAL CODING

INFORMATION DISTORTION AND NEURAL CODING


Alex Dimitrov(1), John Miller(1), Zane Aldworth(1), Tomas Gedeon(2), Albert Parker(2)
(1)Center for Computational Biology and (2)Department of Mathematical Sciences
Montana State University, Bozeman, MT 59717

One of the steps toward understanding the neural basis of an animal's behavior is charactering the code with which its nervous system represents information. We recently used tools from information theory [1] to achieve two goals towards characterizing the neural coding scheme of a simple sensory system as a communication channel. We demonstrated that in this context a coding scheme is an almost deterministic relation between clusters of stimulus/response pairs. Next, we developed a method to find high quality approximations of such a coding scheme. To do this, we quantized the neural responses to a small reproduction set and minimized an information-based distortion function to optimize the quantization.

To use the method in cases involving complex, high dimensional input stimuli, we model the stimulus/response relation in a way that gives us an upper bound to the information distortion, used in the optimization problem. We use it to investigate coding properties of several identified neurons in the cricket cercal sensory system. The results are compared to the linear stimulus reconstruction approach.

[1]AG Dimitrov, JP Miller (2001), Neural coding and decoding: communication channels and quantization, Network: Computation in Neural Systems 12(4), pp441-472