Design-Time Optimization of Deep Neural Networks for Intermittent Learning on Microcontrollers

2026-08-04

AI Summary

This work addresses the challenge of enabling stable on-device deep learning under intermittent power supply—such as solar energy—on resource-constrained microcontrollers, where frequent disruptions hinder reliable training. The authors propose a hardware-aware, offline neural architecture optimization method that, for the first time, integrates multi-objective optimization, automated network design, and intermittent learning in energy-harvesting systems. By constructing an energy prediction model grounded in DNN computational and memory characteristics, the approach accurately estimates layer-wise energy consumption for inference, training, and checkpointing, which is then incorporated into the optimization framework to tailor network architectures to intermittent energy constraints. Experiments on a Cortex-M4 platform using an anomaly detection autoencoder demonstrate a weighted mean absolute percentage error of 16.6% in energy prediction—sufficiently accurate to support robust architectural decisions and enable autonomous edge AI learning.

Abstract

We present a method for designing deep neural networks (DNNs) for intermittent, energy-autonomous, on-device learning on microcontroller units (MCUs). In mobile applications where the energy can run out, e.g., when solar-powered, executing artificial intelligence (AI) faces a technical issue as learning can be interrupted at any time. Our approach combines a hardware-aware energy prediction model with multi-objective optimization (MOO), enabling offline DNN optimization at the design stage without repeated deployment and online testing on the target MCU. Our proposed energy predictor estimates per-layer energy consumption for both DNN inference and training, including the intermittent checkpointing overhead, based on implementation-specific compute and memory features extracted from the DNN model. We validate our approach using autoencoders for anomaly detection on a Cortex-M4 MCU, where our predictor achieves a weighted absolute percentage error of 16.6%, which is sufficient for reliable architecture selection under intermittency constraints. As a result, this work bridges the gap between MOO, automated DNN design, deployment on energy-harvesting systems, and intermittent learning, truly enabling autonomous AI at the edge.

Problem

• intermittent learning
• energy-harvesting systems
• microcontroller units
• deep neural networks
• on-device learning

Innovation

• intermittent learning
• energy prediction
• multi-objective optimization
• microcontroller
• hardware-aware design