Jure Vreča

Journal articlesshow recent

Generating direct logic circuit implementations of deeply quantized neural networks using Chisel4ml, (2025)
Hardware–Software Co-Design of an Audio Feature Extraction Pipeline for Machine Learning Applications, (2024)

Conference papersshow all

Books & Chaptersshow all

Detecting Network Intrusion Using Binarized Neural Networks

J. Vreča, I. Ivanov, G. Papa, A. Biasizzo
Proc. IEEE 7th World Forum on Internet of Things
New Orleans, USA: 14 June-31 July 2021
pages: 622-627
DOI: 10.1109/WF-IoT51360.2021.9595961

BibTex:
@INPROCEEDINGS{VrecaEtAl2021,
author={Vreča, Jure and Ivanov, Iva and Papa, Gregor and Biasizzo, Anton}, booktitle={2021 IEEE 7th World Forum on Internet of Things (WF-IoT)}, title={Detecting Network Intrusion Using Binarized Neural Networks}, year={2021}, volume={}, number={}, pages={622-627}, doi={10.1109/WF-IoT51360.2021.9595961}}

abstract:
The rise of the internet of things will further increase the number of connected devices, which will provide a new attack surface for compromising computer networks. Network intrusion detection systems have already been proposed to help solve this problem, but they are typically computationally too expensive to run on embedded or IoT devices. In this research, we investigate how binarized neural networks can be used to monitor network traffic on such devices, and flag malicious behavior. We train a binarized neural network classifier on the UNSW-NB15 data set and achieve an 82.1% accuracy. We implement the binarized neural network in hardware using a generator written in Chisel and integrate it on a Zynq-7000 system-on-a-chip. We show that such a system can be implemented on even low-end FPGA devices and can increase the security of future networks.