http://terpconnect.umd.edu/~barua/matt-compress-tr.pdf

WK compression algorithms [7] are a unique combination of dictionary and statistical techniques specifically
designed to quickly and efficiently compress program data. To detect and exploit data regularities, the
encoder maintains a dictionary of only 16 words and compresses each input word based on four conditions:
whether the input word is a zero, a word that exactly matches a word in the dictionary, a word that only
partially matches a word in the dictionary (the low bits match and the high bits do not), or a word that
doesn’t match a word in the dictionary. The time complexity of this scheme is similar to LZO but with a
much smaller dictionary size.