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

arXiv: https://arxiv.org/abs/2204.03431 · HTML: https://mlsanity.com/paper/2204.03431/

## Key facts

- Cited by 3 papers in the map (data through 14 Oct 2026).
- Builds on 6 of its 14 references that are in the map.
- First posted on arXiv 7 Apr 2022.

## Builds on (most cited first, 6 of 6)

- [MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications](https://mlsanity.com/paper/1704.04861/) (2017)
- [Speech Commands: A Dataset for Limited-Vocabulary Speech Recognition](https://mlsanity.com/paper/1804.03209/) (2018)
- [PACT: Parameterized Clipping Activation for Quantized Neural Networks](https://mlsanity.com/paper/1805.06085/) (2018)
- [Conditional Deep Learning for Energy-Efficient and Enhanced Pattern Recognition](https://mlsanity.com/paper/1509.08971/) (2015)
- [PULP-NN: Accelerating Quantized Neural Networks on Parallel Ultra-Low-Power RISC-V Processors](https://mlsanity.com/paper/1908.11263/) (2019)
- [Runtime Configurable Deep Neural Networks for Energy-Accuracy Trade-off](https://mlsanity.com/paper/1607.05418/) (2016)

## Built on by (most cited first, 3 of 3)

- [Adaptive Random Forests for Energy-Efficient Inference on Microcontrollers](https://mlsanity.com/paper/2205.13838/) (2022)
- [Energy-Aware Deep Learning on Resource-Constrained Hardware](https://mlsanity.com/paper/2505.12523/) (2025)
- [Accelerating Depthwise Separable Convolutions on Ultra-Low-Power Devices](https://mlsanity.com/paper/2406.12478/) (2024)

## Abstract (arXiv)

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.