Exploring Smartphone-Based Edge AI Inferences Using Real Testbeds

Matías Hirsch ¹, Cristian Mateos ¹, Tim A Majchrzak ^2,3

⁰ISISTAN (UNICEN-CONICET), Tandil 7000, Buenos Aires, Argentina.

Sensors (basel, Switzerland) +

|

May 14, 2025

View abstract on PubMed

Summary

Smartphone clusters offer a competitive edge for AI tasks, providing valuable computing power for real-time applications. This approach enhances edge AI capabilities, especially for computer vision, without heavy cloud reliance.

Area Of Science

Edge Artificial Intelligence (AI)
Computer Vision (CV)
Mobile Computing

Background

Edge AI is expanding due to accessible pre-trained models and AI frameworks.
Deep learning (DL) models are crucial for real-time computer vision tasks like object recognition.
Existing edge AI platforms often rely on cloud resources or homogeneous Single-Board Computers (SBCs), with limited exploration of nomadic hardware like smartphones.

Purpose Of The Study

To investigate the competitiveness of smartphone-based edge AI for real-time computer vision inferences.
To compare the performance of smartphone clusters against SBCs for edge AI workloads.
To evaluate the impact of edge AI tasks on smartphone battery life.

Main Methods

Utilized three pre-trained DL models for computer vision tasks.
Employed eight heterogeneous edge nodes: five low/mid-end smartphones and three SBCs.
Conducted experiments using a toolset for battery-driven edge computing tests across three image stream processing scenarios.

Main Results

Compared latency and energy efficiency between smartphone clusters and SBC-only configurations.
Measured the effect of workload execution on smartphone battery levels in battery-driven settings.
Demonstrated that smartphone clusters can provide significant computing resources for edge AI.

Conclusions

Edge AI leveraging smartphone clusters is a viable and competitive approach for real-time performance.
Smartphone clusters can augment edge AI capabilities, supporting its expansion into diverse application scenarios.
The study provides empirical evidence for the utility of smartphones as edge AI nodes.

Keywords:

cluster computing edge AI energy efficiency smartphones