Complementary information for the paper submitted "MIL in SD"

Abstract

xxxxxxxxxx

 

Empirical study

Dataset domains

We have performed two different studies in this contribution:

  • Firstly, benchmarks that represent two well-known applications in MIL are employed: drug activity and content-based image retrieval. These problems consist of determining whether a drug molecule will bind strongly to a target protein, and the identification of target objects in images, respectively. You can download the datasets in this link.
  • Secondly, we have generated 20 artificial datasets in order to improve the analysis of this study. These datasets have different properties (number of features, instances per bags, and so on) in order to obtain a generalised study. You can download the datasets in this link.

 

Results obtained

Complete results obtained for the CGBA-SD-MIL algorithm:

Complete results obtained for the NMEEFSD-MIL algorithm:

Complete results obtained for the SDMap-MIL algorithm: