[2603.26483] EcoFair: Trustworthy and Energy-Aware Routing for Privacy-Preserving Vertically Partitioned Medical Inference

Computer Science > Machine Learning arXiv:2603.26483 (cs) [Submitted on 27 Mar 2026] Title:EcoFair: Trustworthy and Energy-Aware Routing for Privacy-Preserving Vertically Partitioned Medical Inference Authors:Mostafa Anoosha, Dhavalkumar Thakker, Kuniko Paxton, Koorosh Aslansefat, Bhupesh Kumar Mishra, Baseer Ahmad, Rameez Raja Kureshi View a PDF of the paper titled EcoFair: Trustworthy and Energy-Aware Routing for Privacy-Preserving Vertically Partitioned Medical Inference, by Mostafa Anoosha and 6 other authors View PDF HTML (experimental) Abstract:Privacy-preserving medical inference must balance data locality, diagnostic reliability, and deployment efficiency. This paper presents EcoFair, a simulated vertically partitioned inference framework for dermatological diagnosis in which raw image and tabular data remain local and only modality-specific embeddings are transmitted for server-side multimodal fusion. EcoFair introduces a lightweight-first routing mechanism that selectively activates a heavier image encoder when local uncertainty or metadata-derived clinical risk indicates that additional computation is warranted. The routing decision combines predictive uncertainty, a safe--danger probability gap, and a tabular neurosymbolic risk score derived from patient age and lesion localisation. Experiments on three dermatology benchmarks show that EcoFair can substantially reduce edge-side inference energy in representative model pairings while remaining competitive in cla...