Adaptive Pruning of Deep Neural Networks for Resource-Aware Embedded Intrusion Detection on the Edge
Alexandre Broggi, Nathaniel Bastian, Lance Fiondella, Gokhan Kul
TL;DR
The paper addresses efficient edge deployment of intrusion-detection networks by systematically evaluating pruning methods on the ACI IoT cybersecurity dataset using a fully connected architecture. It compares eight pruning approaches—ADMM-joint, DAIS, Bert-Theseus, ThiNet, Iterative-Theseus, Random pruning, and Recreation—and introduces Iterative-Theseus as a layer-replacement strategy. The findings indicate that most pruning methods do not generalize well to this FC NIDS task, with ThiNet and complete layer-replacement approaches offering the strongest performance and practical pruning times, while DAIS and Bert-Theseus show variability and slower convergence. The study provides guidance for edge deployment by highlighting that model sizing must align with data complexity and that transfer of CNN-centric pruning methods to FC networks is limited.
Abstract
Artificial neural network pruning is a method in which artificial neural network sizes can be reduced while attempting to preserve the predicting capabilities of the network. This is done to make the model smaller or faster during inference time. In this work we analyze the ability of a selection of artificial neural network pruning methods to generalize to a new cybersecurity dataset utilizing a simpler network type than was designed for. We analyze each method using a variety of pruning degrees to best understand how each algorithm responds to the new environment. This has allowed us to determine the most well fit pruning method of those we searched for the task. Unexpectedly, we have found that many of them do not generalize to the problem well, leaving only a few algorithms working to an acceptable degree.