Smoothing Probability Distributions and Smooth Entropy
Christian Cachin and Ueli Maurer
We introduce smooth entropy as a measure for the number of almost uniform random bits that can be extracted from a source by probabilistic algorithms. The extraction process should be universal in the sense that it does not require the distribution of the source to be known. Rather, it should work for all sources with a certain structural property, such as a bound on the maximal probability of any value. The concept of smooth entropy unifies previous work on privacy amplification and entropy smoothing in pseudorandom generation. It enables us to systematically investigate the spoiling knowledge proof technique to obtain lower bounds on smooth entropy and to show new connections to R'enyi entropy of order $\alpha > 1$.