Streaming computation of combinatorial objects

We prove (mostly tight) space lower bounds for "streaming" (or "on-line") computations of four fundamental combinatorial objects: error-correcting codes, universal hash functions, extractors, and dispersers. Streaming computations for these objects are motivated algorithmically by massive data set applications and complexity-theoretically by pseudorandomness and derandomization for space-bounded probabilistic algorithms. Our results reveal a surprising separation of extractors and dispersers in terms of the space required to compute them in the streaming model. While online extractors require space linear in their output length, we construct dispersers that are computable online with exponentially less space. We also present several explicit constructions of online extractors that match the lower bound. We show that online universal and almost-universal hash functions require space linear in their output length (this bound was known previously only for "pure" universal hash functions). Finally, we show that both online encoding and online decoding of error-correcting codes require space proportional to the product of the length of the encoded message and the code\´s relative minimum distance. Block encoding trivially matches the lower bounds for constant rate codes