Finite Gaussian Neurons: Defending Against Adversarial Attacks by Making Neural Networks Say "I Don’t Know"

<p>In this work, I introduce the Finite Gaussian Neuron (FGN), a novel neuron architecture for artificial neural networks aimed at protecting against adversarial attacks.<br />Since 2014, artificial neural networks have been known to be vulnerable to adversarial attacks, which can fool the network into producing wrong or nonsensical outputs by making humanly imperceptible alterations to inputs. While defenses against adversarial attacks have been proposed, they usually involve retraining a new neural network from scratch, a costly task.</p> <p>My works aims to:<br />- easily convert existing models to Finite Gaussian Neuron architecture, <br />- while preserving the existing model's behavior on real data, <br />- and offering resistance against adversarial attacks.</p> <p>I show that converted and retrained Finite Gaussian Neural Networks (FGNN) always have lower confidence (i.e., are not overconfident) in their predictions over randomized and Fast Gradient Sign Method adversarial images when compared to classical neural networks, while maintaining high accuracy and confidence over real MNIST images. <br />To further validate the capacity of Finite Gaussian Neurons to protect from adversarial attacks, I compare the behavior of FGNs to that of Bayesian Neural Networks against both randomized and adversarial images, and show how the behavior of the two architectures differs. <br />Finally I show some limitations of the FGN models by testing them on the more complex SPEECHCOMMANDS task, against the stronger Carlini-Wagner and Projected Gradient Descent adversarial attacks.</p> <p>The code used for this work is available at <a href="https://github.com/grezesf/FGN---Research">https://github.com/grezesf/FGN---Research</a> under the GPL 3.0 open source license. Work done with PyTorch.</p>

Name thesis:degree_name

Doctor of Philosophy

Level thesis:degree_level

Doctoral

Discipline thesis:degree_discipline

Computer Science

Grantor

The Graduate School and University Center of The City University of New York

Year dc:date.available

2022

Author dc:creator

• Grezes, Felix

Advisor dc:contributor.advisor

• Michael I. Mander

Committee members dc:contributor.committeemember

• Rivka Levitan
• Ioannis Stamos
• Andrew Rosenberg

• Artificial Intelligence and Robotics
• Data Science
• Other Computer Sciences
• Machine Learning Adversarial Neural Network

Repository record dc:identifier

https://academicworks.cuny.edu/gc_etds/5129

OAI identifier oai:identifier

oai:academicworks.cuny.edu:gc_etds-6155