The accuracy-fairness-efficiency Trilemma in mobile image classification: a Pareto benchmark

Deploying deep learning classifiers on resource-constrained mobile devices requires simultaneously satisfying three competing objectives: predictive accuracy, demographic fairness, and inference efficiency. No prior work has jointly formalized these as a constrained multi-objective optimization problem, benchmarked a comprehensive strategy set under identical conditions, or defined a Deployment-Feasible Zone (DFZ) as the feasibility-constrained Pareto subset. This paper provides all three contributions. We benchmark eleven optimization configurations on a 2,821-image dataset with 24 intersectional demographic subgroups (worst-case imbalance 35.47:1) under hard constraints ([Formula: see text], [Formula: see text] per attribute, [Formula: see text] MB, [Formula: see text] ms on an entry-level SoC). Three findings emerge. (1) The combination of 3D-aware augmentation and Protected Fairness Pruning (C2) is the Pareto knee point: [Formula: see text] (95 % CI: 0.906-0.962), [Formula: see text], 6.3 MB, 187 ms, confirmed in 94.2 % of bootstrap resamples. (2) Standard magnitude pruning is strictly Pareto-dominated by fairness-constrained pruning (PFP) at identical compression ratio - a result grounded in the low-magnitude minority-encoding effect. (3) The Adaptive Trilemma Weight Scheduler (ATWS) yields consistent gains of [Formula: see text] pp [Formula: see text] and [Formula: see text]-0.7 pp EOD over fixed-weight training, compatible with any fairness-constrained strategy.