Statistics – Machine Learning
Scientific paper
2011-10-28
Statistics
Machine Learning
19 pages
Scientific paper
We introduce a dynamical system which we call the AdaBoost flow. The flow is defined by a system of ODEs with control. We show how by a suitable choice of control the AdaBoost algorithm of Schapire and Freund and the arc-gv algorithm of Breiman can be embedded in the AdaBoost flow. We also show how confidence rated prediction previously studied by Schapire and Singer also can be obtained from our continuous time approach. We introduce a new continuous time algorithm which we call superBoost and describe its properties. The nontrivial part of the AdaBoost flow equations coincides with the equations of dynamics of the nonperiodic Toda system written in terms of spectral variables. This establishes a connection between the two seemingly unrelated fields of boosting algorithms and exactly soluble models of classical mechanics. Finally we explain the similarity of the AdaBoost construction with Perelman's ideas to control the Ricci flow.
Lykov A.
Muzychka S.
Vaninsky K.
No associations
LandOfFree
The AdaBoost Flow does not yet have a rating. At this time, there are no reviews or comments for this scientific paper.
If you have personal experience with The AdaBoost Flow, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and The AdaBoost Flow will most certainly appreciate the feedback.
Profile ID: LFWR-SCP-O-685447