Front Inner Page - Volume 2 No.4 August 2015

  • » Back to Index

  • Title

    :

    Multi Design Picture Analysis Via Huge-Structure Design Interaction Consumed Activity Study

    Authors

    :

    S.Kandhan1, Mr. P.Balamurugan2

    Keywords

    :

    Active learning, multi label classification, high-order label correlation  

    Issue Date

    :

    August – 2015

    Abstract

    :

    Multi-label image classification problems have been solved by the supervised machine learning techniques with incredible profit. In spite of, un-equivalent mechanisms and their performances strongly depend on the premium of training images. However, human annotators have to put significant efforts for the acquisition of training images. This is the major problem in the applications of supervised learning techniques. The iterative learning algorithm has to select the informative example-pairs from where it learns and to learn an accurate classifier with less annotation effort that’s controlled by a high-order label correlation driven active learning (HoAL) approach is proposed in this paper. While, four critical issues are examined by the proposal HoAL. 1) The selection of granularity for the multi-label active learning require to be cleared from example to example-label pair, not alike binary cases. 2) Labels correlations provide critical information for efficient learning and different labels are rarely independent. 3) Additionally, the high-order label correlations and pair-wise label correlations are informative for multi-label active learning. 4) Hence, the number of labels is the reason to increase the number of label combinations exponentially. Whereas to discover the informative label correlations, an efficient mining method is essential. As an empirical result, the demonstration of the proposed approach is effective on public data set. 

    Page(s)

    :

    1-3

    ISSN

    :

    2347- 4734

    Source

    :

    Vol. 2, No.4, August 2015

    Download

    :


  • » Back index