![]() Data compression improves as a side effect of the arms race between eavesdroppers (like the NSA) and steganographers. We would also like to take any function that can distinguish (perhaps probabilistically) between real and randomly generated text and roll it back into making a better compression algorithm. Randomly generated output may be used for art, or more practically, steganography. I'm curious what Burrows-Wheeler and bzip2 will do, since it does not have a probabilistic foundation. First, decompress a training corpus, then generate random text starting from that context state. Motivation was, whereas text compression algorithms have become very sophisticated (Hutter Prize, PAQ family), random sentence generation generally still relies on simple Markov chaining. Disable data integrity checks, or resample until the checks pass. Wherever the decompression (uncompression) program reads input (normally compressed data), let it sample random bits. Modify a data decompressor to produce random output that matches the structural assumptions that allowed data compression to work well in the first place.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |