As a new beginner of the compressiong algorithm, I am interested in studying the high compress ratio algorithm such as PAQ8 and cmix. I find it uses a lot of models to predict each bit in PAQ8, which makes it run slow. After reviewing the code, in the most models, they put the byte contexts (update each byte) in different maps to compute the prediction and update the info in the map by each bit. So I have an idea: if we use the byte contexts to predict next byte directly, it can increase the speed a lot. Different from PPM, we use context models and mixer to increase the accuracy of prediction to differernt types of data.
I am not sure whether there already exists any idea or algorithm about such kinds of byte Prediction (Combing PAQ8 models and Byte Prediction)? And I notice that it may decrease the compression rate but I do not know about how much percent it will decrease?
Thanks