On Iterative Algorithms with an Information Geometry Background

  • Imre Csiszár
Conference paper
Part of the Lecture Notes in Computer Science book series (LNCS, volume 5255)


Several extremum problems in Statistics and Artificial Intelligence, e.g., likelihood maximization, are often solved by iterative algorithms such as iterative scaling or the EM algorithm, admitting an intuitive “geometric” interpretatation as iterated projections in the sense of Kullback information divergence. Such iterative algorithms, including those using Bregman rather than Kullback divergences, will be surveyed. It will be hinted to that the celebrated belief propagation (or sum-product) algorithm may also admit a similar interpretation.

Copyright information

© Springer Berlin Heidelberg 2008

Authors and Affiliations

  • Imre Csiszár
    • 1
  1. 1.Alfréd Rényi Institute of MathematicsHungarian Academy of SciencesBudapestHungary

Personalised recommendations