ATM-Net: Adaptive Termination and Multi-Precision Neural Networks for Energy-Harvested Edge Intelligence

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Solanki, Neeraj, Tabrizchi, Sepehr, Sohrabi, Samin, Schmidt, Jason, Roohi, Arman
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866929715320717312
author Solanki, Neeraj
Tabrizchi, Sepehr
Sohrabi, Samin
Schmidt, Jason
Roohi, Arman
author_facet Solanki, Neeraj
Tabrizchi, Sepehr
Sohrabi, Samin
Schmidt, Jason
Roohi, Arman
contents ATM-Net is a novel neural network architecture tailored for energy-harvested IoT devices, integrating adaptive termination points with multi-precision computing. It dynamically adjusts computational precision (32/8/4-bit) and network depth based on energy availability via early exit points. An energy-aware task scheduler optimizes the energy-accuracy trade-off. Experiments on CIFAR-10, PlantVillage, and TissueMNIST show ATM-Net achieves up to 96.93% accuracy while reducing power consumption by 87.5% with Q4 quantization compared to 32-bit operations. The power-delay product improves from 13.6J to 0.141J for DenseNet-121 and from 10.3J to 0.106J for ResNet-18, demonstrating its suitability for energy-harvesting systems.
format Preprint
id arxiv_https___arxiv_org_abs_2502_09822
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle ATM-Net: Adaptive Termination and Multi-Precision Neural Networks for Energy-Harvested Edge Intelligence
Solanki, Neeraj
Tabrizchi, Sepehr
Sohrabi, Samin
Schmidt, Jason
Roohi, Arman
Machine Learning
ATM-Net is a novel neural network architecture tailored for energy-harvested IoT devices, integrating adaptive termination points with multi-precision computing. It dynamically adjusts computational precision (32/8/4-bit) and network depth based on energy availability via early exit points. An energy-aware task scheduler optimizes the energy-accuracy trade-off. Experiments on CIFAR-10, PlantVillage, and TissueMNIST show ATM-Net achieves up to 96.93% accuracy while reducing power consumption by 87.5% with Q4 quantization compared to 32-bit operations. The power-delay product improves from 13.6J to 0.141J for DenseNet-121 and from 10.3J to 0.106J for ResNet-18, demonstrating its suitability for energy-harvesting systems.
title ATM-Net: Adaptive Termination and Multi-Precision Neural Networks for Energy-Harvested Edge Intelligence
topic Machine Learning
url https://arxiv.org/abs/2502.09822