Energy-efficient machine learning models that can run directly on edge devices are of great interest in IoT applications, as they can reduce network pressure and response latency, and improve privacy. An effective way to obtain energy-efficiency with small accuracy drops is to sequentially execute a set of increasingly complex models, early-stopping the procedure for 'easy' inputs that can be confidently classified by the smallest models. As a stopping criterion, current methods employ a single threshold on the output probabilities produced by each model. In this work, we show that such a criterion is sub-optimal for datasets that include classes of different complexity, and we demonstrate a more general approach based on per-classes thresholds. With experiments on a low-power end-node, we show that our method can significantly reduce the energy consumption compared to the single-threshold approach.
Energy-efficient adaptive machine learning on IoT end-nodes with class-dependent confidence / Daghero F.; Burrello A.; Pagliari D.J.; Benini L.; Macii E.; Poncino M.. - ELETTRONICO. - (2020), pp. 9294863.1-9294863.4. (Intervento presentato al convegno 27th IEEE International Conference on Electronics, Circuits and Systems, ICECS 2020 tenutosi a gbr nel 2020) [10.1109/ICECS49266.2020.9294863].
Energy-efficient adaptive machine learning on IoT end-nodes with class-dependent confidence
Burrello A.;Benini L.;
2020
Abstract
Energy-efficient machine learning models that can run directly on edge devices are of great interest in IoT applications, as they can reduce network pressure and response latency, and improve privacy. An effective way to obtain energy-efficiency with small accuracy drops is to sequentially execute a set of increasingly complex models, early-stopping the procedure for 'easy' inputs that can be confidently classified by the smallest models. As a stopping criterion, current methods employ a single threshold on the output probabilities produced by each model. In this work, we show that such a criterion is sub-optimal for datasets that include classes of different complexity, and we demonstrate a more general approach based on per-classes thresholds. With experiments on a low-power end-node, we show that our method can significantly reduce the energy consumption compared to the single-threshold approach.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
