Paper #198

Title:

The minimax distortion redundancy in empirical quantizer design

Authors:

Peter Bartlett, Tamas Linder and Gábor Lugosi

Date:

January 1997

Abstract:

We obtain minimax lower and upper bounds for the expected distortion redundancy of empirically designed vector quantizers. We show that the mean squared distortion of a vector quantizer designed from $n$ i.i.d. data points using any design algorithm is at least $\Omega (n^{-1/2})$ away from the optimal distortion for some distribution on a bounded subset of ${\cal R}^d$. Together with existing upper bounds this result shows that the minimax distortion redundancy for empirical quantizer design, as a function of the size of the training data, is asymptotically on the order of $n^{1/2}$. We also derive a new upper bound for the performance of the empirically optimal quantizer.

Keywords:

Estimation, hypothesis testing, statistical decision theory: operations research

JEL codes:

C13, C14

Area of Research:

Statistics, Econometrics and Quantitative Methods

Published in:

IEEE Transactions on Information Theory, 44, (1998), pp. 1802-1813

Download the paper in PDF format