CompLearn: Apply Compression to Discover Patterns

I first heard about this in the context of language categorization.

You compress samples of each language and then to categorize a new sample you compress it with each compressor and the one that does the best job tells you what language.