Template:Using Data Compressors for Robust Reasoning
From What The Wiki?!
Using Data Compressors for Robust Reasoning
For four decades, the field of data compression has attracted some ofthe world's best statistical minds. Using basic principles of statisticalreasoning, we have managed to create a wide variety of "general" purposedata compressors such a bzip, gzip, and PPM that can shorten files. In mytalk I will explain how to use these same programs to do all sorts ofinteresting pattern recognition. I will present a way to convert anycompression program into a statistical inference engine suitable for awide variety of tasks, including:a) evolutionary tree reconstruction from genomic sequence datab) detecting viruses, worms or other bad packetsc) determining language relationships from text samplesIn more recent research, we have recast the Google search engine as asort of pseudo-compressor, and used this to use Google to do veryadvanced semantic classifications, so that a computer can by itselfdetermine the difference between colors and numbers, or arrange paintingsby painters, without ever being told explicitly what to do. This possibilityto do objective or subjective reasoning holds great promise for futuremachine learning applications.
