The New Jersey Institute of Technology's
Electronic Theses & Dissertations Project

Title:	Towards adversarial robustness with 01 lossmodels, and novel convolutional neural netsystems for ultrasound images
Author:	Xie, Meiyan
View Online:	njit-etd2021-051 (xiv, 81 pages ~ 4.9 MB pdf)
Department:	Department of Computer Science
Degree:	Doctor of Philosophy
Program:	Computer Science
Document Type:	Dissertation
Advisory Committee:	Roshan, Usman W. (Committee chair) Wei, Zhi (Committee member) Neamtiu, Iulian (Committee member) Ady, Justin (Committee member) Graves, William (Committee member)
Date:	2021-08
Keywords:	Adversarial robust Deep learning Machine learning Medical AI Plaque segmentation Vessel segmentation
Availability:	Unrestricted
Abstract:	This dissertation investigates adversarial robustness with 01 loss models and a novel convolutional neural net systems for vascular ultrasound images. In the first part, the dissertation presents stochastic coordinate descent for 01 loss and its sensitivity to adversarial attacks. The study here suggests that 01 loss may be more resilient to adversarial attacks than the hinge loss and further work is required. In the second part, this dissertation proposes sign activation network with a novel gradient-free stochastic coordinate descent algorithm and its ensembling model. The study here finds that the ensembling model gives a high minimum distortion (as measured by HopSkipJump) compared to full precision, binary, and convolutional neural networks, and explains this phenomenon by measuring the transferability between networks in an ensemble. In the last part, this dissertation tackles three important segmentation problems for vascular ultrasound images with novel convolutional neural networks. More specifically, these three problems are: (1) vessel segmentation in the internal carotid artery, (2) vessel segmentation in the entire carotid system, and (3) vessel and plaque segmentation in the entire carotid system. The study here represents a first successful step towards the automated segmentation of vessel and plaque in carotid artery ultrasound images and is an important step in creating a system that can independently evaluate carotid ultrasounds.