| The advancement in semiconductor, fiber optics, sensor, and wireless technologies enabled a wide variety of new applications as well as improvements in the existing ones. While some of these advancements provide abundant resources of a certain type, efficient and optimized data representation is still a crucial system parameter that has large impact on overall system performance. More generally, compression is a way to optimize data representation in order to increase computational and communication capabilities of the existing infrastructure.;The focus of our attention is the modeling for lossless compression. Nowadays the most successful lossless compression schemes utilize sophisticated adaptive statistical modelling of input stream in order to predict upcoming symbols and feeding the prediction to a coder. In order to improve the statistical modeling, we have developed a technique for the cleaning of already collected information about the input stream such that its most important properties are emphasized even more.;As the next step of data modeling we introduce the compression technique for program binaries. The technique efficiently combines the analysis of binaries with this general purpose compressor. We introduce novel modeling for the lossless compression of an input stream by adapting the binary for better compression. We leverage on the fact that functionally equivalent versions of the same binary have different compression ratios, and use it to modify binaries in such a way that their compression ratios are improved.;A next step towards a higher level of abstraction of modeling data for compression is compression aware design, which includes modified and new data representations such that they facilitate the usage for initial purposes as well as better compression. We have demonstrated the effectiveness of data modeling on this level of abstraction with two examples in the area of wireless ad hoc networks: (i) compression of routing tables in shortest path routing, and (ii) trajectory data compression.;All three levels of abstraction of data modeling of compression enables better compression. Nevertheless, presented techniques enable new and exciting areas of research for improving not only the compression ratio, but also the compression/decompression speed, and data characterization via compression. |
