Federated learning and tiny models improve IoT device security
Reliable Federated TinyML Deployment for IoT Security
Cryptography and SecurityMachine Learning
Summary
IoT devices often have limited computing power but need to detect security threats while keeping data private. The paper studies how to combine a privacy-friendly training method called federated learning with small, efficient AI models that can run on tiny devices. The authors find that certain training tricks, especially a learning rate schedule controlled by a server, make the models much better at spotting attacks even after being made smaller and faster. This work helps design lightweight security tools for small connected devices without sharing sensitive data.
What this means in practice
- •For iot device manufacturers: Implement efficient and privacy-aware intrusion detection models that run directly on constrained IoT hardware.$Commercial implications: Enables development of secure embedded systems that detect attacks locally without exposing raw data, suitable for commercial IoT devices.
- •For network security teams: Deploy improved federated TinyML systems to enhance attack detection accuracy on distributed edge devices with limited resources.
Authors
Younsoo Park, Seokhyoen Bae, Shasi Kumar Ramachandran Prabhu, Suman Saha, Peilong Li
Abstract
The growing deployment of Internet of Things (IoT) devices has increased the need for privacy-preserving intrusion detection systems that operate directly on resource-constrained hardware. Federated Learning enables collaborative model training without sharing raw data, but conventional federated models are often too large and unstable for deployment on microcontroller-class devices. TinyML techniques enable compact neural networks but are typically designed for inference-only workloads. This work investigates combining Federated Learning with TinyML-based model compression for intrusion detection in IoT environments. We evaluate compression strategies including knowledge distillation, structured pruning, and quantization within a federated training pipeline. Preliminary results show that training stability plays a critical role in federated TinyML systems. In particular, server-coordinated cosine learning-rate scheduling improves Attack Recall from 46.7% to 93.85% while enabling substantial model compression and efficient edge deployment. These findings provide insights for designing lightweight and privacy preserving intrusion detection systems for IoT devices.