Metacognitive ensemble method speeds up mobile sensor data analysis
Metacognitive Selective Ensemble for Mobile Systems
Machine Learning
Summary
Mobile devices use multiple AI models working together to better understand sensor data, but running all models continuously wastes battery and memory. The authors developed MetaSE, a smart way to pick only the best models at each moment, avoiding running every model all the time. MetaSE remembers which models worked well recently and quickly swaps them out when needed, saving time and memory without losing accuracy. Tests showed it matches the accuracy of running all models while being much faster and lighter.
What this means in practice
- •For mobile app developers: Improve energy and memory efficiency of sensor-based apps by selectively running only reliable models at a time without sacrificing accuracy.
- •For embedded system engineers: Deploy accurate human activity recognition with less computational load on constrained devices like Raspberry Pi using adaptive model ensembles.
Authors
Sungmin Lee, Kichang Lee, Joonhee Lee, JaeYeon Park, Songkuk Kim, JeongGil Ko
Abstract
Deep ensembles improve robustness in mobile sensing, but repeatedly executing many models over continuous sensor streams is costly. Selecting only a few members reduces this cost, yet adaptive selection often requires additional model execution to obtain reliable evidence about inactive candidates. We present MetaSE, an active ensemble framework that exploits short-term persistence in per-model reliability. MetaSE maintains a small active set across windows, uses post-execution evidence to reject unreliable members, and invokes lightweight routing only when replacement is needed. This stateful design accesses the diversity of a larger pool without repeated full-pool evaluation. Across four HAR datasets and four model architectures, MetaSE consistently improves over a fixed three-model ensemble and achieves accuracy comparable to substantially more expensive adaptive and full-ensemble inference. On a Raspberry Pi 4B, MetaSE is 2.7x faster and uses 69% less memory than full ten-model inference.