On reliability of content identification from databases based on noisy queries

Error exponents for the relay channel

“…We can optimize for the conditional distribution QŶ 2 |Y2,X2 explaining the inner maximization over QŶ 2 |Y2,X2 and outer minimization over Q Y2|X2 . This is a game-theoretic-type result similar to [10]- [12].…”

“…For CF, since it is related to Wyner-Ziv, we leverage on the work of Kelly-Wagner [10] who derived an achievable exponent for Wyner-Ziv. In a similar vein, [11] and [12] derived lower bounds for the error exponents of Gel'fand-Pinsker and content identification respectively. We also note that Ngo et al [13] presented an achievable error exponent for amplify-forward for the AWGN relay channel but does not take into account block-Markov coding.…”

“…Hence, all that remains is to bound the probability of the union in (12). There are two cases: For the case where the decoded bin indexk j is correct (i.e., equal to K j ) but message ism j wrong (i.e., not equal to 1), the analysis is relatively straightforward.…”

Reliability of Universal Decoding Based on Vector-Quantized Codewords

“…For a memoryless source and channel, the maximum achievable rate R (i.e., the capacity) of this model setting has already been established by Tuncel [11] (see also [13], [14], [15]). Two years later, Dasarathy and Draper [2] have derived a lower bound to the achievable reliability (achievable error exponent) at a given rate R, and then after three more years [3], the same authors have also derived an upper (converse) bound to the reliability function based on a sphere-packing argument.…”

“…In this paper, we improve on the analysis in [2]. In particular, while Dasarathy and Draper chose to analyze the performance of the well-known maximum mutual information (MMI) decoder [1], without an apparent explanation and justification for this choice of decoder, here we argue that, in this special setting, there is room for improvement over the MMI decoder, in two different aspects.…”