Evaluation of Jensen-Shannon distance over sparse data

Connor, Richard and Cardillo, Franco Alberto and Moss, Robert and Rabitti, Fausto (2013) Evaluation of Jensen-Shannon distance over sparse data. In: Similarity Search and Applications. Lecture Notes in Computer Science, 8199 . Springer, Berlin, pp. 163-168. ISBN 9783642410611

[img]
Preview
 Text (Connor-etal-LNCS2013-Evaluation-Jensen-Shannon-distance-over-sparse-data)
Connor_etal_LNCS2013_Evaluation_Jensen_Shannon_distance_over_sparse_data.pdf
Accepted Author Manuscript
Download (249kB)| Preview

    Abstract

    Jensen-Shannon divergence is a symmetrised, smoothed version of Küllback-Leibler. It has been shown to be the square of a proper distance metric, and has other properties which make it an excellent choice for many high-dimensional spaces in R*. The metric as defined is however expensive to evaluate. In sparse spaces over many dimensions the Intrinsic Dimensionality of the metric space is typically very high, making similarity-based indexing ineffectual. Exhaustive searching over large data collections may be infeasible. Using a property that allows the distance to be evaluated from only those dimensions which are non-zero in both arguments, and through the identification of a threshold function, we show that the cost of the function can be dramatically reduced.

    Author(s):Connor, Richard ORCID logoORCID: https://orcid.org/0000-0003-4734-8103, Cardillo, Franco Alberto, Moss, Robert and Rabitti, Fausto
    Item type:Book Section
    ID code:52372
    Keywords:distance metrics, exhaustive searching, high dimensional spaces, instrinsic dimensionalitites, Jensen-Shannon divergence, metric spaces, other properties, threshold functions, artificial intelligence, computer science, Electronic computers. Computer science, Information Systems
    Subjects:Science > Mathematics > Electronic computers. Computer science
    Department:Faculty of Science > Computer and Information Sciences
    Depositing user: Pure Administrator
    Date deposited:31 Mar 2015 07:23
    Last modified:15 Nov 2019 04:24
    Related URLs:
    URI:https://strathprints.strath.ac.uk/id/eprint/52372
    Export data:
    CORE (COnnecting REpositories)