On-line learning using Semi-Supervised Learning (SSL) and Transduction as New Learning Technology for Undergraduate Bioinformatics Studies

Presenter: Nada Basit

Time: May 11, 3:30-4:45

Location: B122

Bioinformatics is a relatively young field of study. Its goal is to increase the understanding of biological processes. It is the application of statistics and computer science to the field of molecular biology. Bioinformatics focuses on developing and applying computationally intensive techniques (such as machine learning algorithms, pattern recognition, data mining, and visualization) to achieve this goal. Standard “one-shot” machine learning techniques (e.g. Semi-Supervised Learning (SSL) alone) based upon induction and deduction may not be the most appropriate methods for Bioinformatics studies because biological data is mostly unlabeled with only small amounts of labeled data (labeled with respect to a class.) To address this, a Statistical Learning Theory (SLT) approach using Transduction in conjunction with SSL is proposed for the new learning requirements. Additionally, justification for using an “on-line” approach versus standard “one-shot” learning will be discussed. Finally, an example comparing the standard one-shot SSL technique to the new on-line trandsductive SSL approach will be shown.

Comments are closed.