25th Annual Symposium onFoundations of Computer Science, 1984. (1984)
Singer Island, FL
Oct. 24, 1984 to Oct. 26, 1984
A. El Gamal , Stanford University
Let X and Y be two random variables with probability distribution p(x,y), joint entropy H(X,Y) and conditional entropies H(X \ Y) and H(Y \ X) . Person P/sub x/ knows X and person P/sub Y/ knows Y. They communicate over a noiseless two-way channel so that both know X and Y. It is proved that, on the average, at least H(X \ Y) + H(Y \ X) bits must be exchanged and that H(X,Y) + 2 bits are sufficient. If p(x.y) > 0 for all (x.y), then at least H(X,Y) bits must be communicated on the average. However, if p (x,y) is uniform over its support set, the average number of bits needed is close to H(X \ Y) + H (Y \ X). Randomized protocols can reduce the amount of communication considerably but only when some probability of error is acceptable.
A. El Gamal and A. Orlitsky, "Interactive Data Compression," 25th Annual Symposium onFoundations of Computer Science, 1984.(FOCS), Singer Island, FL, 1984, pp. 100-108.