Special Presentation: Pablo Zegers

    Tuesday, January 27, 2015 - 1:30pm
    Franken Conference Room (Meinel 821)

    "Fisher Information Properties"


    Fisher information and Shannon entropy measure system order. This is from opposite directions: Whereas increased system order increases the Fisher information, it decreases the Shannon entropy. An introduction to current uses of information theory is given, in particular in application to the detection of binary stars. The need to broaden the scope of use of these techniques becomes evident. Next, a set of Fisher information properties is presented in order to draw a parallel with similar properties of Shannon entropy. These include already-known properties, along with new ones that include generalizations of the Kullback-Leibler entropy and Shannon mutual information. These quantities are used to prove that Fisher information: i) increases under conditioning, ii) decreases for Markov chains and iii) forces upper bounds on errors of estimation.

    Speaker Bio(s): 

    Pablo Zegers is a professor of electrical and computer engineering at the University of the Andes in Santiago, Chile.