Identification over multiple databases

The tradeoff between storage and identification rates for multiple databases is investigated from an information theoretic perspective. In the assumed model, noisy observations of feature vectors of two distinct groups, called the ancestors, are compressed and stored in two separate databases. When queried with a noisy observation of a (possibly random) function of two randomly selected ancestors (one from each group), the system is required to correctly identify the ancestors with high probability. Single-letter inner and outer bounds are presented on the set of achievable rate points, which identify a tradeoff between the compression rates and the identification rate region: the lower the compression rates for storage, the larger the rate region achievable for identification.