Abstract
Online learning machines are essential tool for real-time adaptive systems
such as autonomous robots.
However, if the input data include a large number of unrelated variables, the
learning machine wastes a high computational power to complete the
learning.
This is because, the learning machine needs a huge number of samples to
detect which variables are related to the desired outputs precisely.
To overcome this problem, we propose a quick online learning methods
using speculative filter and wrapper methods.
In this method, the system achieves quick online dimension selection and
learning even if the input samples are independent distributed samples.
