CIS Master's Thesis Defense by Alexandre Broggi
Date: May 6, 2025
Time and Location: 11am in Dion 311
Thesis Advisor: Dr. Gokhan Kul, CIS Faculty
Committee members:
Dr. Yuchou Chang, CIS Faculty
Dr. Firas Khatib, CIS Faculty
Dr. Jiawei Yuan, CIS Faculty
Thesis Title: Application and Ontological Exploration of Artificial Neural Network Pruning for Network Intrusion Detection
Abstract:
Artificial Neural Networks are an increasingly popular way to approach solutions against real world problems. They have been applied to a myriad of problems in transportation, medicine and cybersecurity. A significant problem that has been approached with Neural Networks in the past few years, is network activity evaluation. However, there are some difficulties using Neural Networks in real-world systems due to the fact that online activities generating too large workloads resulting in overly large and complex models to use, where it becomes necessary to either generate smaller models or prune them to a more desirable size.
This paper attempts to create a simple Artificial Neural Network detection system for network activity, specifically from network packets, and prune the model created through a variety of techniques, and then analyze the model to try to gain an understanding of the effects that pruning has on the model using an ontological approach. This is different from previous studies in three key ways: (i) the application of ontological methods on the model itself, instead of on a related system, (ii) the wide array of pruning methods used for analysis, and (iii) application on Network Intrusion Detection systems to make it usable in the real-world.
Our findings can be used to further the path of the exploration of artificial neural network pruning by posing some new paths of inquiry and closing off some algorithms that were not able to function well. The specific ontological methods can add information for how much further a Neural Network can be reliably pruned, as well as finding which methods work best on the Network Intrusion Detection problem. For reproducibility and transparency purposes, we contribute our codebase for further use and exploration.
For further information please contact Dr. Gokhan Kul
Dion 311