PAQ
PAQ is a family of context-mixing compressors developed by Matt Mahoney and others that repeatedly set compression ratio records at extremely slow speeds.
Purpose
It pursues the best possible compression ratio (the Hutter Prize), achieving the highest known ratios on text and mixed data.
Mechanism
It integrates many models (context models, literal matches, WordModel, and dedicated image/audio models), mixes their predictions with a neural network or logistic mixer, and outputs through an arithmetic coder.
Use cases
- Compression ratio competitions
- Maximum-ratio archiving
- Compression algorithm research