1: \begin{abstract}

2: This paper  introduces a new notion of dimensionality of probabilistic models from an information-theoretic view point.

3: We call it the {\em descriptive dimension}~(Ddim).

4: We show that Ddim coincides with the number of independent  parameters for the parametric class, and can further be extended to real-valued dimensionality when a number of models are mixed.

5: The paper then derives the rate of convergence of the MDL (Minimum Description Length) learning algorithm which outputs a normalized maximum likelihood (NML) distribution with model of the shortest NML codelength.

6: The paper proves that the rate is governed by Ddim.

7: The paper also derives error probabilities of the  MDL-based test for multiple model change detection. It proves that they  are also governed by Ddim.

8: Through the analysis, we demonstrate that Ddim is an intrinsic quantity which characterizes the  performance of the MDL-based learning and change detection.

9: 

10: \end{abstract}

11: