Engineering Commons^™

Gradient Free Sign Activation Zero One Loss Neural Networks For Adversarially Robust Classification, Yunzhe Xue

Dissertations

The zero-one loss function is less sensitive to outliers than convex surrogate losses such as hinge and cross-entropy. However, as a non-convex function, it has a large number of local minima, andits undifferentiable attribute makes it impossible to use backpropagation, a method widely used in training current state-of-the-art neural networks. When zero-one loss is applied to deep neural networks, the entire training process becomes challenging. On the other hand, a massive non-unique solution probably also brings different decision boundaries when optimizing zero-one loss, making it possible to fight against transferable adversarial examples, which is a common weakness in deep learning …

Towards Adversarial Robustness With 01 Lossmodels, And Novel Convolutional Neural Netsystems For Ultrasound Images, Meiyan Xie

Dissertations

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 …

Towards Understanding The Role Of Central Processing In Release From Masking, Nima Alamatsaz

Dissertations

People with normal hearing have the ability to listen to a desired target sound while filtering out unwanted sounds in the background. However, most patients with hearing impairment struggle in noisy environments, a perceptual deficit which current hearing aids and cochlear implants cannot resolve. Even though peripheral dysfunction of the ears undoubtedly contribute to this deficit, surmounting evidence has implicated central processing in the inability to detect sounds in background noise. Therefore, it is essential to better understand the underlying neural mechanisms by which target sounds are dissociated from competing maskers. This research focuses on two phenomena that help suppress …

Drug Releasing Hydrogels For Opioid Use Disorder, Kaytlyn M. Crowe

Theses

Ease of access to prescription opioids and the strength of synthetic opioids have contributed to the rise in over use disorders, overdose, and death rates during the opioid epidemic. The current overdose standard of care, rescue NLX (NLX), is a competitive antagonist to the μ-opioid receptor but has a relatively short period of action compared to opioid agonists, especially synthetic agonists. Rises in the abuse of these synthetic and semi-synthetic agents in recent years have shown the weaknesses of rescue NLX, as it can leave overdose patients vulnerable to renarcotization and precipitated withdrawal. To prevent both from occurring, we developed …

Alcohol As A Catalyst For Hiv-Associated Neuroinflammation And Tbi-Induced Iron Toxicity, Agnieszka Agas

Dissertations

Alcohol has long been considered an exacerbator of diseases, disorders, and injuries as well as many of the accompanying symptoms. As an alternative approach, this dissertation explores alcohol as a catalyst for two different human disease conditions, human immunodeficiency virus (HIV)-associated neuroinflammation and traumatic brain injury (TBI)-induced iron toxicity. In HIV-1 infection, this dissertation presents a novel anti-viral drug, called Drug-S, for a possible inhibition and treatment of HIV-1 disease progression.

The first aim explores the influence of alcohol with HIV-associated neuroinflammation on macrophage migration across an in vitro model of the blood brain barrier. There is a gap in …

Development Of Deep Learning Neural Network For Ecological And Medical Images, Shaobo Liu

Dissertations

Deep learning in computer vision and image processing has attracted attentions from various fields including ecology and medical image. Ecologists are interested in finding an effective model structure to classify different species. Tradition deep learning model use a convolutional neural network, such as LeNet, AlexNet, VGG models, residual neural network, and inception models, are first used on classifying bee wing and butterfly datasets. However, insufficient data sample and unbalanced samples in each class have caused a poor accuracy. To make improvement the test accuracy, data augmentation and transfer learning are applied. Recently developed deep learning framework based on mathematical morphology …

Using Fnirs As An Objective Measure Of Susceptibility To Informational Masking, Min Zhang

Dissertations

Resolving complicated auditory scenes is crucial for daily communication where background sound is often present. However, most hearing aid (HA) and cochlear implant (CI) users have difficulties understanding speech when competing sound sources are present, resulting in reduced job opportunities and increased risk for social isolation. Perceptual interference from background sound is called auditory masking. At least two distinct masking phenomena exist, called energetic and informational masking (EM and IM).

In the first masking phenomenon, EM, target and masker energies coincide at the same time and frequency. Computational and physiological models of cochlear auditory processing can reliably predict listeners' performances …

The Effects Of Design Parameters On The Neural Recordings With Micro-Ecog Arrays, Manan Amish Sethia

Theses

In the field of neural prosthetics, electro-cortico-graph (ECoG) arrays are commonly used to record neural activity of the brain cortex both in animal and human subjects. A finite element model (FEM) was developed to simulate the electric field generated by a single neuron in the rat brain cortex and a micro ECoG array (µECoG) placed on the pia surface for recording the neural signal. The neuron was simulated as a dipole current source with a magnitude of 1µA and placed at three different depths in the motor cortex corresponding to different layers under the µECoG array. The array design was …

Real Time Analysis Of Eye Movements Using Computer Aided Software, Farhan Ahmad

Theses

Eye movement analysis is a slow and tasking job that require a trained vision analyst to sift through hundreds of eye movements. The difficulty of the task increases when a subject has undiagnosed eye related disorders. One such disorder is Convergence insufficiency (CI) which affects nearly 12 million people in the US. Convergence insufficiency occurs when the eyes are unable to coordinate simultaneous inward movement for near focal tracking. Individuals with CI are unable to coordinate eye movements this can result in many symptoms such as double vision, blurred vision, headaches, vision strain, and fatigue. A diagnosis for CI relies …