<ul><li>The demand for machine learning model training on edge devices is increasing due to data privacy and personalized service needs.</li><li>A two-stage data selection framework called Titan is proposed to optimize data resource utilization for on-device model training.</li><li>Titan filters out important data batches in the first stage and uses an optimal data selection strategy in the second stage for improved model performance.</li><li>Empirical results show that Titan reduces training time by up to 43% and increases final accuracy by 6.2% on edge devices with minor system overhead.</li></ul>

A Two-Stage Data Selection Framework for Data-Efficient Model Training on Edge Devices

Discover more