Bayesian Feedback Analytics for Quantum-Assisted Target Recovery in Security-Critical Data Spaces
Main article
Abstract
Grover's search and its amplitude-amplification generalisation give a quadratic speed-up for locating marked entries in an unstructured data space, yet recovering every marked entry rather than a single one introduces a fragile dependence on the assumed number of targets. Because the optimal number of amplification rounds is fixed by this count, an estimate that drifts from the truth either halts the search early—leaving targets undiscovered—or drives it into an unbounded hunt for entries that do not exist. We frame this difficulty as an online inference problem and present a quantum-classical procedure that treats the unknown target count as a random variable, maintaining a posterior over candidate counts that is revised after each measurement through Bayesian feedback analytics. A lightweight classical layer prunes improbable hypotheses, defers updates until the recent outcome record is informative, and protects neighbourhood diversity so that the true count is not discarded prematurely. Across search spaces from 2^12 to 2^24 the method drives the mean number of unrecovered targets close to zero—for a space of 2^24 entries the static baseline left roughly 60.9 targets unfound on average while our procedure left 0.82—and it does so without inflating cost, requiring about 389,400 oracle calls against 408,000 for the baseline. A vulnerability-enumeration case study on large open-source code bases shows the same pattern, recovering all catalogued defects where the fixed-count baseline missed several percent. We argue that adaptive posterior correction should be a default component of any complete-enumeration quantum search in security-critical settings
