Application and Analysis of Expectation Propagation Algorithm in the Binary Erasure Quantization
share
Regarding the widespread utilization of big data, compression and reconstruction of valuable data sets get more essential from the aspect of data storage management. The Binary Erasure Quantization (BEQ) problem considers data-worth for compression and is solved by a dual form of the Belief Propagation (BP) algorithm. Nevertheless, the dual of BP does not converge to the optimum performance because of adverse short cycles in the associated graph of the compression code. We propose a form of Expectation Propagation (EP) algorithm, which gives rise to a precise inference for the BEQ. By the suggested algorithm, the empirical compression rate with zero distortion approximately tends to the theoretical BEQ lossless compression rate.
READ FULL TEXT
