Information encoding in sensory systems

Abstract

Environmental signals sensed by the nervous system are often represented in spike trains carried from sensory neurons to higher neural functions where decisions and functional actions occur. Information about the environmental stimulus is thought to be encoded in these spike trains. We are trying to understand the dynamically processes involved in representing a continuous stimulus as a series of discrete spiking events. We have been using the well known visual neuron in the flesh fly (Sarcophaga Bullata) called the H1 to study these processes. This neuron is thought to transmit information about the velocity of horizontal motion across the whole visual field. The fly then uses this information to calculate its rotational velocity for use in flight course control.

In order to understand how patterns of correlated spikes can be used to encode information we have applied the state space methods of nonlinear dynamics to reading the neural code. Using these methods we can generate the mapping from a series of consecutive spikes to the input driving the neuron. This mapping is realized locally in a reconstructed state space embodying both the dynamics of the source of the sensory signal and the dynamics of the neural circuit doing the processing. We show that one may accurately learn the dynamical input/output connection and estimate with high precision the details of the input signals from spike timing output alone. Knowing this mapping is equivalent to "reading" the neural code. This form of "reading the neural code'' has a focus on the neural circuitry as a dynamical system and emphasizes how one interprets the dynamical degrees of freedom in the neural circuit as they transform analog environmental information into spike trains. We have explored this idea using a Hodgkin-Huxley conductance based neuron model and have started experiments with the H1.

Studies have shown that the H1 can reliably generate identical responses to repeated presentations of a stimulus when the stimulus signal-to-noise ratio is high (i.e. the motion stimulus is bright). This implies that a particular stimulus is encoded by a particular spike train. Of course this does not imply that the mapping be one-to-one; meaning that two distinct stimuli could produce the same output spike train. But it does seem to imply that the encoding is a deterministic process. Currently we are trying to use this reliability in the spike timing to study general properties of the encoding process in the H1. We have built a rig that can deliver a bright, wide-field motion stimulus to the fly. We present a particular stimulus many times and record the spike times. Then we make smooth change to the stimulus and repeat that motion many times and record those spike times. For the H1 the stimulus is defined by the angular velocity of a visual pattern that we move in front of the fly. Therefore a smooth change of the stimulus could involve scaling the amplitude of the angular velocity or the time scale of the motion presentation. Since for each stimulus the spikes tend to occur at the same times on each repetition, we can try to compare the neuronal response to the original stimulus with the response to the scaled stimulus. The goal is to look for a scaling relationship in the input/output relationship of the H1 encoding process.

People

E. Tumer
J. Wolfe
H. D. I. Abarbanel

Publications


Abarbanel HDI, Masuda N, Rabinovich MI, Tumer, E
Distribution of mutual information
Phys Lett A 281(5-6):368-373 (2001)

Abarbanel HDI and Tumer E
Reading Neural Encodings Using Phase Space Methods
in "Perspectives and Problems in Nonlinear Science :  A Celebratory Volume in Honor of Lawrence Sirovich"
Kaplan E, Marsden JE, and Sreenivasan KR(eds.)
Springer 2003
Comments? Contact
Terry Peters, Phone +1-858-534-7753, tpeters (at) ucsd.edu