The information in the compressed file is sufficient to reconstruct the original file. This is done with the ``uncompress'' algorithm, which works as follows: again, for each character , the predictor (sequentially) emits its output based on the previous characters, where the with were obtained sequentially by feeding the approximations of the probabilities into the inverse Huffman Coding procedure. The latter is able to correctly decode from . Note that to correctly decode some character, we first need to decode all previous characters.

Juergen Schmidhuber 2003-02-19