The accuracy–fairness–efficiency Trilemma in mobile image classification: a Pareto benchmark
Thanh TranVan, Hien LamThanh, Toan DoNang, Tuan-Tu Huynh · 2026
Read The accuracy–fairness–efficiency Trilemma in mobile image classification: a Pareto benchmark by Thanh TranVan, Hien LamThanh, Toan DoNang, Tuan-Tu Huynh (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/pmc-13478527.txt.
Description
Thanh TranVan and three colleagues benchmarked in 2026 the three-way trade-off between accuracy, fairness, and speed in mobile image classifiers. Phones must run small, fast models that still treat demographic groups evenly, yet improving one goal usually harms another.
They tested eleven tuning setups on 2,821 images covering 24 demographic subgroups under strict size and speed caps. A mix of 3D-aware data augmentation and fairness-aware pruning hit the sweet spot at 6.3 MB and 187 ms, holding up in 94.2 percent of resamples. Plain magnitude pruning lost to fairness-aware pruning at the same compression, and an adaptive weight scheduler added small steady gains. The work maps a feasible zone where mobile models can be accurate, fair, and fast at once.
Catalog record
| CREATOR | Thanh TranVan, Hien LamThanh, Toan DoNang, Tuan-Tu Huynh |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 48 KB |
| VERSION | 1.0 |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |