Energy-Constrained Deployment of Learning Models on Distributed Research Sensors: A Cross-Site Validation Agenda

Abstract

Edge sensors must trade the convenience of local AI inference against limited battery life and intermittent connections. The difficulty is particularly visible when a system is transferred to a new setting. The central research objective is to identify operating conditions in which local prediction is preferable to delayed cloud processing. Previous work provides context for sampling, input validation and study boundaries [3–7]. To answer the question, the study proposes a discrete-event experiment using simulated energy budgets and repeatable sensor sampling patterns. Related enterprise-storage studies inform the assessment of operational reliability [1] and the deployment assumptions adopted for system-level comparison [2]. Under matched evaluation conditions, the protocol compares continuous edge inference, compressed models and scheduled cloud offloading. Methodological sources are used to specify transparent comparison and reporting practices [8–12]. The cross-site extension holds out an entire institution or deployment location rather than relying only on random test splits. Performance will be judged using energy per valid inference, end-to-end delay, missed anomalies and buffering requirements. Interpretation will account for connectivity losses, battery ageing, temperature effects and uncertainty in simulated demand. Between-site heterogeneity would be examined through prespecified subgroup estimates and transparent domain-shift diagnostics. The supplementary technical literature is used to identify plausible failure mechanisms [13–17]. No study result is assumed in advance; the analysis is designed to show both when the method helps and when it fails, with reporting context from the remaining sources [18–23].