Distributed K-means over Compressed Binary Data
Résumé
We consider a network of sensors which transmit their measurements to a fusion center which has to perform K-means clustering on the received data. The sensors compress their data with an LDPC code, and we propose to apply the K-means algorithm directly over the compressed data without reconstructing the original measurements. From a theoretical analysis, we show that it is reasonable to apply the K-means algorithm in the compressed domain, and we design the LDPC code parameters in order to optimize the performance of the K-means algorithm. At the end, we show from Monte Carlo simulations that the rate needed to perform K-means clustering in the compressed domain is lower than the rate needed to reconstruct all the measurements.