Citation

Tan, Jun Wei and Goh, Pey Yun and Tan, Shing Chiang and Chong, Lee Ying (2024) Projecting the Pattern of Adversarial Attack in Low Dimensional Space. In: 2024 12th International Conference on Information and Communication Technology (ICoICT), 07-08 August 2024, Bandung, Indonesia.

Text Projecting the Pattern of Adversarial Attack in Low Dimensional Space.pdf - Published Version Restricted to Repository staff only Download (1MB)

Abstract

One of the serious cyber dangers to machine learning (ML) are the adversarial attacks. Since deep learning neural network (DNN) flourished in processing all the images data, this newly advanced ML has been ‘harassed’ by adversarial attacks. Significant failure can be seen when image data is perturbed even a little. Researchers have attempted to understand the pattern and identify the characteristics of advrsarial attacks. In this study, we try to project the pattern of adversarial attacks in a way that can be understood by both machines and humans. Thus, uniform manifold approximation and projection (UMAP) is applied to project the attacks from high dimensional space to low dimensional space. However, as UMAP may not efficient in reducing all kind of data and thus, we introduced another model where feature extraction layers of a convolution neural network (CNN) are utilized. This model is known as DeepUMAP. Three publicly available datasets and three adversarial attacks are applied to compare the performance between UMAP and DeepUMAP through graph visualization. We also employed k-means and a clustering index to describe the efficacy of UMAP and DeepUMAP. Encouraging results are reported with DeepUMAP.

Item Type:	Conference or Workshop Item (Paper)
Uncontrolled Keywords:	convolution neural network, dimension reduction, adversarial attacks
Subjects:	Q Science > QA Mathematics > QA71-90 Instruments and machines > QA75.5-76.95 Electronic computers. Computer science
Divisions:	Faculty of Information Science and Technology (FIST)
Depositing User:	Ms Nurul Iqtiani Ahmad
Date Deposited:	04 Dec 2024 03:12
Last Modified:	04 Dec 2024 03:12
URII:	http://shdl.mmu.edu.my/id/eprint/13220

Downloads

Edit (login required)