Selecting salient features for machine learning from large candidate pools through parallel decision-tree construction

摘要

parallel algorithms, protein folding 1 Introduction The primary goal of machine learning research is to develop algorithms that enable computers to learn to perform various tasks. These can range anywhere from guiding a robot's motion through complex environments to predicting tomorrow's weather. The common goal which all machine learning systems share is that of substituting learning for explicit programming, in the hope that this may lead to easier, faster, and even better system construction than is possible through hand coding. A major category of machine learning systems is that of inductive classification systems. Such systems learn to categorize observational instances, or examples, in some sensible way, either according to a predefined set of classifications (supervised learning) or a set of freely discovered regularities in the examples (unsupervised learning). This chapter is mainly concerned with inductive classification systems within the paradigm of supervised learning from examples.