Transfer Learning Methods and Applications in Computational Biology

Gunnar RätschMPG.PuRe 2009

Abstract

I will present and discuss methods and applications of transfer learning in computational biology. Here, one brings together knowledge that have been obtained in experiments on various organisms to understand the differences and similarities of biological processes in related organisms. I will start with an empirical comparison of several methods for transfer learning in a sequence classification setting, where the domains correspond to organisms. Moreover, I will present new approaches for multitask learning where the tasks have a hierarchical relationship. We were particularly interested in how to make the learning algorithms scalable and how to effectively use the knowledge about the relations between the tasks for learning. I will show applications of these techniques in a biological sequence classification problem with 15 organisms and preliminary results for transfer learning in a structured output prediction setting.
BibTeX
@inproceedings{ratsch2009transfer,
  author = {Gunnar Rätsch},
  title = {{Transfer Learning Methods and Applications in Computational Biology}},
  booktitle = {MPG.PuRe},
  year = {2009},
}

Loading related content...

Figure