Abstract:Advances in Tiny Machine Learning (TinyML) have bolstered the creation of smart industry solutions, including smart agriculture, healthcare and smart cities. Whilst related research contributes to enabling TinyML solutions on constrained hardware, there is a need to amplify real-world applications by optimising energy consumption in battery-powered systems. The work presented extends and contributes to TinyML research by optimising battery-powered image-based anomaly detection Internet of Things (IoT) systems. Whilst previous work in this area has yielded the capabilities of on-device inferencing and training, there has yet to be an investigation into optimising the management of such capabilities using machine learning approaches, such as Reinforcement Learning (RL), to improve the deployment battery life of such systems. Using modelled simulations, the battery life effects of an RL algorithm are benchmarked against static and dynamic optimisation approaches, with the foundation laid for a hardware benchmark to follow. It is shown that using RL within a TinyML-enabled IoT system to optimise the system operations, including cloud anomaly processing and on-device training, yields an improved battery life of 22.86% and 10.86% compared to static and dynamic optimisation approaches respectively. The proposed solution can be deployed to resource-constrained hardware, given its low memory footprint of 800 B, which could be further reduced. This further facilitates the real-world deployment of such systems, including key sectors such as smart agriculture.

What problem does this paper attempt to address?

The paper primarily addresses the energy optimization problem in battery-powered TinyML (Tiny Machine Learning) systems for image anomaly detection. The core objective of the research is to optimize the operational strategies of these resource-constrained systems through Reinforcement Learning (RL) methods to extend their deployment duration. Specifically, the paper explores the following key points: 1. **Background and Challenges**: With the development of TinyML technology, more and more application scenarios are being realized, such as smart agriculture, healthcare, and smart cities. However, in these application scenarios, devices typically rely on limited battery power and need to perform complex tasks such as image processing and anomaly detection. This leads to a new challenge—how to effectively manage the limited energy supply while ensuring the functionality of the devices. 2. **Research Contributions**: The paper proposes a scheme that uses reinforcement learning algorithms to optimize battery-powered TinyML systems, particularly for image anomaly detection tasks. Through simulation experiments, it is verified that compared to static and dynamic optimization methods, this scheme can significantly improve the battery life of the devices, increasing it by 22.86% and 10.86%, respectively. 3. **Optimization Methods**: The paper compares three methods: static optimization, dynamic optimization, and autonomous optimization based on reinforcement learning. The autonomous optimization method employs a Decayed Epsilon-Greedy Q-learning algorithm, which enables the device to self-learn the optimal operational strategies, including when to upload anomaly images to the cloud for processing and when to perform model training locally. 4. **Simulation Environment**: To verify the effectiveness of the scheme, the researchers built a simulation environment to test the battery life performance under different optimization strategies. The simulation environment considers various factors, such as changes in anomaly detection rates, the choice of wireless communication methods (e.g., NB-IoT), and the energy consumption required for model training. 5. **Results Analysis**: The experimental results show that the autonomous optimization scheme based on reinforcement learning can significantly enhance battery life, especially under changing environmental conditions. Additionally, the scheme has good portability and adaptability, making it suitable for resource-constrained hardware platforms and capable of continuously learning and adapting to environmental changes during actual deployment. In summary, this paper provides an effective energy management strategy that helps improve the feasibility and practicality of TinyML systems in real-world deployments.

Simulating Battery-Powered TinyML Systems Optimised using Reinforcement Learning in Image-Based Anomaly Detection

Explore Training of Deep Convolutional Neural Networks on Battery-powered Mobile Devices: Design and Application

An Efficient and Flexible Learning Framework for Dynamic Power and Thermal Co-Management

Enhancing Battery Management System Optimization: A Secure AI Modeling Approach

Is TinyML Sustainable? Assessing the Environmental Impacts of Machine Learning on Microcontrollers

Optimizing TinyML: The Impact of Reduced Data Acquisition Rates for Time Series Classification on Microcontrollers

TinyML for Ultra-Low Power AI and Large Scale IoT Deployments: A Systematic Review

On-device Online Learning and Semantic Management of TinyML Systems

TinyML: Tools, Applications, Challenges, and Future Research Directions

RIOT-ML: toolkit for over-the-air secure updates and performance evaluation of TinyML models

Power Efficient Machine Learning Models Deployment on Edge IoT Devices

Optimal Computation Resource Allocation in Energy-Efficient Edge IoT Systems with Deep Reinforcement Learning

On TinyML and Cybersecurity: Electric Vehicle Charging Infrastructure Use Case

How to Manage Tiny Machine Learning at Scale: An Industrial Perspective

Energy Efficient Predictive Maintenance in Resource Constraint Industrial IoT devices using Smoothed Mish Activation Function

Reinforcement Learning based Multi-Access Control and Battery Prediction with Energy Harvesting in IoT Systems

Introduction to the Special Issue on tinyML

Maximizing the Performance of a Lithium-Ion Battery Aging Estimator Using Reinforcement Learning

Efficient Neural Networks for Tiny Machine Learning: A Comprehensive Review

Adaptive Resolution Inference (ARI): Energy-Efficient Machine Learning for Internet of Things