Daniel Wichs: Extracting Randomness from Exractor-Dependent Sources

Abstract: We revisit the well-studied problem of extracting nearly uniform randomness from an arbitrary source of sufficient min-entropy. Strong seeded extractors solve this problem by relying on a public random seed, which is unknown to the source. Here, we consider a setting where the seed is reused over time  and the source may depend on prior calls to the extractor with the same seed. Can we still extract nearly uniform randomness? In more detail, we assume the seed is chosen randomly, but the source can make arbitrary oracle queries to the extractor with the given seed before outputting a sample. We require that the sample has entropy and differs from any of the previously queried values. The extracted output should look uniform even to a distinguisher that gets the seed. We  provide constructions and lower bounds for extractors in this setting.