Wednesday 05 March 2025
The Internet of Things (IoT) has brought about a new era of connectivity, where devices of all shapes and sizes are constantly communicating with each other. However, this increased connectivity also brings new challenges in terms of security and device management. One of the most pressing issues is identifying and tracking IoT devices, particularly when it comes to firmware updates.
In recent years, researchers have been working on developing techniques for identifying and classifying IoT devices based on their network traffic patterns. These approaches typically involve analyzing the packet flows and protocol usage of the devices to determine their type, make, and model. However, these methods often rely on specific traffic patterns and may not be effective in situations where devices are using encryption or other obfuscation techniques.
A new approach has been proposed by researchers that uses a technique called transfer learning with twin neural networks (TNNs) to identify IoT devices based on their firmware versions. The idea is to use a neural network trained on a dataset of known device types and firmware versions to generate a fingerprint of each device’s traffic patterns. This fingerprint can then be used to identify unknown devices and determine their firmware version.
The researchers tested their approach using data from 12 IoT devices, including smart home appliances, cameras, and routers. They found that the TNNs were able to accurately identify devices based on their firmware versions with an accuracy rate of over 95%. They also demonstrated that the approach was effective even when faced with encrypted traffic or devices using obfuscation techniques.
One of the key benefits of this approach is its ability to handle unknown devices. Unlike traditional approaches that rely on specific traffic patterns, TNNs can learn to recognize new devices and firmware versions without requiring additional training data. This makes it an attractive solution for IoT device management and security applications where devices may be constantly changing or updating.
The researchers also explored the use of Hedges’ g effect size to measure the similarity between devices and identify subtle changes in traffic patterns that may indicate a firmware update. They found that this approach was able to accurately detect changes in firmware versions even when they were subtle, demonstrating its potential for real-world applications.
Overall, the proposed approach offers a promising solution for IoT device identification and management. By leveraging transfer learning with TNNs, it is possible to develop systems that can effectively identify and track devices based on their firmware versions, even in situations where traditional approaches may struggle.
Cite this article: “Identifying IoT Devices Using Transfer Learning with Twin Neural Networks”, The Science Archive, 2025.
Iot, Device Identification, Transfer Learning, Twin Neural Networks, Tnns, Firmware Updates, Network Traffic Patterns, Encryption, Obfuscation Techniques, Device Management
