Multiscale Bootstrap Method and Its Applications

Abstract

How much confidence we may have in a discovery from data? This problem becomes evident insome applications of Bioinformatics. A novel algorithm for measuring the randomness in data is shown; itcalculates a highly accurate unbiased probability value in terms of statistical hypothesis testing. Geometry,such as distance and curvature, of the space of probability distributions plays an important role in thealgorithm. The key idea is rescaling of the space, which is analogous to the renormalization theory ofstatistical physics.