Combining Semi-Supervised Learning and Relational Resampling for Active Learning in Network Domains

Download Now Free registration required

Executive Summary

Recent work in statistical relational learning has demonstrated the effectiveness of network-based classification methods, which exploit relational dependencies among in-stances to improve predictions. These methods have been applied in a broad range of domains, from bioinformatics to fraud detection. Although labeled training examples can be costly to acquire in these domains, there has been little work focusing on active learning techniques that can identify the most beneficial (network) instances to label for learning. Past work has mainly focused on learning with a xed set of labeled nodes| either from a fully labeled or partially labelled network.

  • Format: PDF
  • Size: 353 KB