REDUNDANCY REDUCTION IN THE ASCENDING AUDITORY PATHWAY 

  G. Chechik1; M.J. Anderson4; E.D. Young4; I. Nelken1,2*; N. Tishby1,3 

  1. Interdisciplinary Center for Neural Computation, Hebrew
     University, Jerusalem, Israel
  2. Physiology, Hebrew University-Hadassah Medical School, Jerusalem,
     Israel
  3. School of computer science and engineering, Hebrew University,
     Jerusalem, Israel
  4. Biomedical Engineering, Johns Hopkins University, Baltimore, MD, USA


  The way complex acoustic signals are transformed along the auditory
  system is still largely unknown. To address this question we have
  recorded the neural activity of neurons in both primary auditory
  cortex (A1) and inferior colliculus (IC) of halothane anesthetized
  cats in response to natural and modified bird songs. We estimated
  the amount of information about the stimulus carried by single and
  small groups of neurons first as the information provided by total
  spike counts on stimulus identity, and secondly as the information
  provided by single spikes on the acoustic signals that preceded
  them.  Spike counts of single IC neurons provided high information
  about the stimulus, but were significantly redundant. In contrast,
  single A1 neurons carried about half the information but were
  largely independent. Thus, information could be summed over small
  groups of cortical neurons. In fact, using a simple neural network
  allowed a reliable readout of the stimulus using 10-20 cortical
  cells. Weak dependencies between A1 cells, in comparison with high
  IC cells correlations, were also demonstrated in the coding of the
  acoustics with single spikes. These results are in agreement with
  the hypothesis that an important aspect of processing along
  ascending sensory systems is redundancy reduction. The reduced
  redundancy may allow easier readout of relevant information from
  small sets of neurons.  

  Supported by: Human Frontier Science Project(HFSP) grant RG
  0133/1998 and the Israeli ministry of science.