Energy-Efficient Adaptive Machine Learning on IoT End-Nodes With Class-Dependent Confidence

TL;DR

This paper introduces a class-dependent confidence threshold method for energy-efficient machine learning on IoT end-nodes, improving upon single-threshold approaches by reducing energy consumption while maintaining accuracy.

Contribution

The work proposes a novel per-class threshold approach for early stopping in adaptive models, enhancing energy efficiency on IoT devices with diverse class complexities.

Findings

Significant energy savings compared to single-threshold methods

Effective early stopping for easy inputs reduces computation

Method maintains accuracy across different class complexities

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.