**** Columbia Network Research Center Seminar ****

Title:    Photo to Grandma Problem: Compression Meets the Network 
Speaker:  Jelena Kovacevic 
		  Bell Labs, Lucent Technologies
		  Murray Hill, NJ
When:     December 13, 2000 - 10 am
Where:    Interschool Lab, CEPSR Building, 7th floor
          Columbia University


Abstract:

  In communication, separating source coding from channel coding is a
  standard form of modularization.  It makes things simple for us, and
  furthermore Shannon's Separation Principle allows us to feel that we
  are not going to suffer for it.  For practical purposes, the existence
  of coding theorems with separate source and channel characterizations
  is much less important than the fact that separation leads to big
  toolboxes of reusable tools.  However, we should not confuse the
  sophistication of existing tools with their appropriateness.

  We'll examine certain communication scenarios and draw conclusion on
  the existing toolboxes.  I'll show that there is "beyond"
  multiresolution.  In particular, in some communication scenarios, the
  information available at the source decoder is a subset of a small
  number of chunks of data.  (Think of an unreliable packet network
  without retransmission of lost packets.) The right tools for source
  coding are not the conventional ones, but rather multiple description
  (MD) codes.  We'll conclude with demonstrations of MD speech and audio
  coders.