On-device learning based vulnerability detection in IoT environment

Pre-trained machine learning models have demonstrated significant potential for enhancing vulnerability detection in Internet of Things (IoT) systems. By applying quantization techniques, this work constrained model weights and activations to binary values to significantly reduce model size and computational cost, making them deployable on IoT devices. A novel Binary Neural Networks (BNN) scheme is proposed to binarize vulnerability detection models, optimizing memory usage and computational costs on resource-constrained IoT devices. A vulnerability detection BNN was implemented with optimized parameters, including the number of activation layers, kernel size, and the size of the fully connected layer. The BNN model was evaluated on a Raspberry Pi using the IoT23 and NSL-KDD datasets, demonstrating promising performance in vulnerability detection.