Compressing human text.

  • Christian Steinruecken

I will present a new data compression algorithm that compresses human text more effectively than all current variants of PPM, CTW, DMC, LZ (Lempel-Ziv), LZMA, CSE and BWT. It uses a hierarchical non-parametric sequence model and online gradient optimisation.