Applying model selection to statistical sequence modeling: An MDL interpretation of the PPM compression algorithm

This work aims to create a bridge between statistical sequence modeling research, represented by the PPM compression algorithm, with model se- lection, represented by the Minimum Description Length Principle (MDL). By creating a direct link between the sequential prediction strategy of PPM with the inductive inference theory for selecting models of MDL, we see that PPM can be seen as a type of a conditional universal distribution relative to a variable-order Markov model. Making this connection allows future research to bridge the gap between PPM and MDL literature.