Review of Large-Scale Coordinate Descent Algorithms for Multi-class Classification with Memory Constraints

Abstract

Big data can impact the performance of many standard measures used for classification. Specifically the efficiency of multi-class classification algorithms when the dataset is too large to fit into limited memory available needs to be explored. Different algorithms with varying complexity have been proposed in the literature. Two of the most recognized classification algorithms, batch learning and online learning have emerged as the most consistent options when solving for a multi-class problems. Presently a gap in the documentation of such algorithms exists in the literature available. Furthermore, the recent development of the online multi-class solver warrants a detailed examination. This thesis will address both concerns, providing detailed documentation of the analysis, comparisons of the algorithms, plots of the results, as well as a discussion about the findings.