Presentation Title: Evaluation and Comparison of Adaptive Sampling Methods for Learning by Boosting

Committee:

• Dr. Jianhua Chen (chair)
• Dr. Sitharama Iyengar
• Dr. Konstantin Busch

Date: Apr 27, 2011
Time: 11:00 AM
Location:256 Coates Hall

Abstract:
In machine learning research, we often need to handle tasks of learning from large dataset. Huge dataset poses computational challenges to learning algorithms because the large data size demands efficient learning methods otherwise the computational time for getting the required results could be infeasible. Sampling can be used to alleviate the computational burden in learning when the dataset is huge. The “Madaboost” method by Watanabe is a boosting algorithm that uses the concept of adaptive sampling for learning by boosting. In this MS project, we have experimentally evaluated our new adaptive sampling method with several benchmark datasets from the UCI machine learning repository. Also, we have compared our results with the results of Watanabe’s Madaboost method. We have compared these two adaptive sampling techniques in terms of running time efficiency and prediction accuracy performance with different datasets. The evaluation results show that our method is much more efficient and produces competitive or better prediction accuracy.


All are invited.