Answers are grounded in this paper and cite their sources.
Cross-building transfer learning can reduce Wi-Fi fingerprint calibration effort, but it may degrade positioning when source and target environments have incompatible access-point deployments, propagation conditions, or spatial mappings. This paper proposes SafeTransLoc, a framework that treats transfer as an optional deployment decision: target-only and transfer candidates are compared on held-out target data, and transfer is accepted only when its validation advantage exceeds a safety margin. The principal UJIIndoorLoc and SODIndoorLoc evaluation comprises 120 cross-building configurations; under the deployment-level criterion, the best transfer candidate was worse than the strongest target-only alternative in all of them. A post-review controlled same-environment SODIndoorLoc experiment adds 270 RSS-offset/noise configurations and yields 6 beneficial-transfer and 264 harmful-transfer cases. With a zero margin, the gate accepted 1 of the 6 beneficial cases and rejected 89.0% of harmful transfers; increasing the margin improved harmful-transfer rejection but reduced beneficial-transfer sensitivity. Independent leave-one-environment-out calibration on the archived SOD results selected 0 m for CETC331 and 0.5 m for HCXY and SYL, producing one held-out false acceptance overall. These results support SafeTransLoc as a deployment safeguard while showing that conservative validation gating involves a measurable safety-sensitivity trade-off under sparse target supervision.
No takes yet. Share an insight, caveat, or question.
Sa’ahiry et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: