Theses and Dissertations from UMD
Permanent URI for this communityhttp://hdl.handle.net/1903/2
New submissions to the thesis/dissertation collections are added automatically as they are received from the Graduate School. Currently, the Graduate School deposits all theses and dissertations from a given semester after the official graduation date. This means that there may be up to a 4 month delay in the appearance of a give thesis/dissertation in DRUM
More information is available at Theses and Dissertations at University of Maryland Libraries.
Browse
8 results
Search Results
Item Dynamical Memory in Deep Neural Networks -(2024) Evanusa, Matthew S; Aloimonos, Yiannis; Computer Science; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)In this work, I will begin to lay out a roadmap or framework for which I believe will serve the scientific communities of artificial intelligence and cognitive neuroscience of interest, in future development and design of a thinking intelligent machine, based on the accumulated knowledge I have gathered across many sources: from my advisors, peers and colleagues, collaborators, talks, symposia and conferences, and long paper dives, for the almost decade that I have spent at my new home in College Park, Maryland. It is my hope and intent that this thesis serves in its stated goal to advance the science of memory integration in neural networks, but in addition, to further the distant dream of discovering the mystery of what it means to be alive. It is important to note that while this thesis is focused on the critical integration of memory mechanisms into artificial neural networks, the authors’ larger goal is the creation of an overarching cognitive architecture that takes advantages of the right amount of advances from deep learning, with the right amount of insights from cognitive and neuroscience - a ”Goldilocks” of sorts for AI. It is my hope that through understanding mechanisms of memory and how they interact with our stimluli, we move one step closer to understanding our place in the cosmos.Item Adversarial Robustness and Fairness in Deep Learning(2023) Cherepanova, Valeriia; Goldstein, Tom; Applied Mathematics and Scientific Computation; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)While deep learning has led to remarkable advancements across various domains, the widespread adoption of neural network models has brought forth significant challenges such as vulnerability to adversarial attacks and model unfairness. These challenges have profound implications for privacy, security, and societal impact, requiring thorough investigation and development of effective mitigation strategies. In this work we address both these challenges. We study adversarial robustness of deep learning models and explore defense mechanisms against poisoning attacks. We also explore the sources of algorithmic bias and evaluate existing bias mitigation strategies in neural networks. Through this work, we aim to contribute to the understanding and enhancement of both adversarial robustness and fairness of deep learning systems.Item Affective Human Motion Detection and Synthesis(2022) Bhattacharya, Uttaran; Manocha, Dinesh; Computer Science; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)Human emotion perception is an integral component of intelligent systems currently being designed for a wide range of socio-cultural applications, including behavior prediction, social robotics, medical therapy and rehabilitation, surveillance, and animation of digital characters in multimedia. Human observers perceive emotions from a number of cues or modalities, including faces, speech, and body expressions. Studies in affective computing indicate that emotions perceived from body expressions are extremely consistent across observers because humans tend to have less conscious control over their body expressions. Our research focuses on this aspect of emotion perception as we attempt to build predictive methods for automated emotion recognition from body expressions, and build generative methods for synthesizing digital characters with appropriate affective body expressions. This thesis elaborates on both components of our research in two parts, and explores how they can be applied to current problems in video understanding, specifically video highlight detection. The first part discusses two approaches for designing and training partially supervised methods for emotion recognition from body expressions, specifically gaits. In one approach, we leverage existing gait datasets annotated with emotions to generate large-scale synthetic gaits corresponding to the emotion labels. In the other approach, we leverage large-scale unlabeled gait datasets together with smaller annotated gait datasets to learn meaningful latent representations for emotion recognition. We design an autoencoder coupled with a classifier to learn latent representations for simultaneously reconstructing all input gaits and classifying the labeled gaits into emotion classes. The second part discusses generative methods to synthesize emotionally expressive bodily expressions, specifically gaits, gestures, and faces. The first method involves asynchronous generation, where we synthesize only one modality of the digital characters (in our case, gaits) with affective expressions. Our approach is to design an autoregression network that takes in a history of the characters' pose sequences and the intended future emotions to generate their future pose sequences with the desired affective expressions. The second method is the more challenging synchronous generation, where the affective contents of two modalities, such as body gestures and speech, need to be synchronized with each other. Our approach utilizes machine translation to translate from speech to body gestures, and adversarial discrimination to differentiate between original and synthesized gestures in terms of affective expressions, to produce state-of-the-art affective body gestures synchronized with speech. The final method takes synchronous generation a step further to three modalities, involving the synthesis of both facial expressions and body gestures synchronized with speech. This method attempts to break new ground in multimodal synthesis by simultaneously incorporating emotional expressions in more than one modality, and does so using data from affordable, consumer-grade devices such as RGB video cameras to enable democratized usage. Lastly, we explore the application of these approaches to industrial problems in video understanding, specifically video highlight detection. Our approach leads to state-of-the-art performance in detecting highlights in human-centric videos without requiring supervision in the form of highlight annotations. Our approach can be further fine-tuned to detect user-specific highlights at scale by automatically learning the video contents matching the users' preferences in their previously selected highlight clips.Item TOWARDS AN EFFICIENT SEMANTIC SEGMENTATION PIPELINE FOR 3D ELECTRON MICROSCOPY DATA.(2022) Emam, Zeyad Ali Sami; Czaja, Wojciech; Goldstein, Thomas; Applied Mathematics and Scientific Computation; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)In recent years, deep neural networks revolutionized many aspects of computer vision. However, their success relies on massive high-quality annotated datasets that are costly to curate. This thesis is composed of three major parts. In Chapter 3, we use novel high dimensional visualization methods to explore connections between the loss landscape of neural networks and their intriguing ability to generalize to unseen test data. Next, in Chapter 4, we tackle a difficult computer vision task, namely the segmentation of anisotropic 3D electron microscopy image volumes. Deep neural networks tend to struggle in this scenario due to the lack of sufficient training data and the 3 dimensional nature of the images, as such we develop a novel state-of-the-art architecture and training workflow to improve the overall segmentation pipeline. Finally, in Chapter 5 we propose a novel state-of-the-art deep active learning algorithm for image classification to alleviate the costs of data annotations and allow networks to train effectively using less data.Item Spintronics-based Architectures for non-von Neumann Computing(2020) Mondal, Ankit; Srivastava, Ankur; Electrical Engineering; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)The scaling of transistor technology in the last few decades has significantly impacted our lives. It has given birth to different kinds of computational workloads which are becoming increasingly relevant. Some of the most prominent examples are Machine Learning based tasks such as image classification and pattern recognition which use Deep Neural Networks that are highly computation and memory-intensive. The traditional and general-purpose architectures that we use today typically exhibit high energy and latency on such computations. This, and the apparent end of Moore's law of scaling, has got researchers into looking for devices beyond CMOS and for computational paradigms that are non-conventional. In this dissertation, we focus on a spintronic device, the Magnetic Tunnel Junction (MTJ), which has demonstrated potential as cache and embedded memory. We look into how the MTJ can be used beyond memory and deployed in various non-conventional and non-von Neumann architectures for accelerating computations or making them energy efficient. First, we investigate into Stochastic Computing (SC) and show how MTJs can be used to build energy-efficient Neural Network (NN) hardware in this domain. SC is primarily bit-serial computing which requires simple logic gates for arithmetic operations. We explore the use of MTJs as Stochastic Number Generators (SNG) by exploiting their probabilistic switching characteristics and propose an energy-efficient MTJ-SNG. It is deployed as part of an NN hardware implemented in the SC domain. Its characteristics allow for achieving further energy efficiency through NN weight approximation, towards which we develop an optimization problem. Next, we turn our attention to analog computing and propose a method for training of analog Neural Network hardware. We consider a resistive MTJ crossbar architecture for representing an NN layer since it is capable of in-memory computing and performs matrix-vector multiplications with O(1) time complexity. We propose the on-chip training of the NN crossbar since, first, it can leverage the parallelism in the crossbar to perform weight update, second, it allows to take into account the device variations, and third, it enables avoiding large sneak currents in transistor-less crossbars which can cause undesired weight changes. Lastly, we propose an MTJ-based non-von Neumann hardware platform for solving combinatorial optimization problems since they are NP-hard. We adopt the Ising model for encoding such problems and solving them with simulated annealing. We let MTJs represent Ising units, design a scalable circuit capable of performing Ising computations and develop a reconfigurable architecture to which any NP-hard problem can be mapped. We also suggest methods to take into account the non-idealities present in the proposed hardware.Item Integrated Field Inversion and Machine Learning With Embedded Neural Network Training for Turbulence Modeling(2019) Holland, Jonathan Richard; Baeder, James D; Duraisamy, Karthik; Aerospace Engineering; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)A rich set of experimental and high fidelity simulation data is available to improve Reynolds Averaged Navier Stokes (RANS) models of turbulent flow. In practice, using this data is difficult, as measured quantities cannot be used to improve models directly. The Field Inversion and Machine Learning (FIML) approach addressed this challenge through an inference step, in the form of an inverse problem, which treats inconsistencies between the models and the data in a consistent manner. However, a separate learning algorithm is not always able to be learned from the generated inverse problem data accurately. Two new methods of incorporating higher fidelity data into RANS turbulence models via machine learning are proposed and applied for the first time in this thesis. Both build on the FIML framework by performing learning during the inference step, instead of considering the inference and learning steps separately as in the classic FIML approach. The first new approach embeds neural network learning into the RANS solver, and the second trains the weights of the neural network directly. Additionally, for the first time, the inverse problem can incorporate higher fidelity data from multiple cases simultaneously, promising to improve the generalization of the augmented model. The two new methods and the classic approach are demonstrated with a simple model problem, as well as a number of challenging RANS cases. For a 2D airfoil case, all three FIML augmentations are shown to improve predictions, with the new methods demonstrating increased regularization. Additionally, a model augmentation is generated by considering seven angles of attack of an airfoil in the inference step, and the augmentation is shown to improve predictions on a different airfoil. Additional cases are considered including a transonic shock wave boundary layer interaction and the NASA wall-mounted hump. In all cases, the inference is shown to improve predictions. For the first time, the inverse problem accounts for the limitations of the learning procedure, guaranteeing that the model discrepancy is optimal for the chosen learning algorithm. The results in this thesis prove that learning during the inference step provides additional regularization, and guarantees the inference produces learnable model discrepancy.Item ESTIMATING SURFACE LONGWAVE RADIATION AND APPLICATIONS TO HIGH LATITUDE ISSUES(2012) Nussbaumer, Eric; Pinker, Rachel T; Atmospheric and Oceanic Sciences; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)Two models, with distinct advantages for calculating downwelling surface longwave (DSLW) radiation under all sky conditions are presented. Both models are driven with a combination of Moderate Resolution Imaging Spectroradiometer (MODIS) level-3 cloud parameters and information from the European Centre for Medium-Range Weather Forecasts (ECMWF) ERA-Interim model. To compute the clear sky component of DSLW the first model DSLW/UMD v1 utilizes a globally applicable parameterization. The second generation model DSLW/UMD v2 utilizes a two layer feed-forward artificial neural network with sigmoid hidden neurons and linear output neurons. When computing the cloud contribution to DSLW, DSLW/UMD v1 implements a commonly used statistical model to calculate cloud vertical height while in DSLW/UMD v2 the cloud base temperature is estimated by using an independent artificial neural network based on spatially and temporally co- located MODIS and Cloudsat Cloud Profiling Radar (CPR) and the Cloud-Aerosol Lidar and Infrared Pathfiner Satellite Observation (CALIPSO) Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP) observations. Daily average estimates of DSLW for 2003 to 2009 are compared against ground measurements from the Baseline Surface Radiation Network (BSRN) and show significant improvements over currently available model estimates. DSLW/UMD v2 as optimized for Polar Regions along with a UMD develop shortwave model are used to investigate the role of radiative components in Arctic sea ice anomalies. The correlation between downwelling surface longwave and shortwave radiation and sea ice anomaly for the period from 2003 to 2007 is investigated using the latest Moderate Resolution Imagining Spectroradiometer (MODIS) level-3 cloud parameters and information from the European Centre for Medium-Range Weather Forecasts (ECMWF) ERA-Interim model. All sky downwelling surface longwave radiation (DSLW), all sky downwelling shortwave radiation (DSSW), all sky total downwelling shortwave and longwave radiation (DSSW + DSLW), and cloud total cloud forcing are individually examined to determine their respective correlation to sea ice anomaly. It is determined that these radiation components are not the primary drivers for major sea ice anomalies that occur during the investigated time frame within the 120o E to 210o E region.Item Design of Beam Steering Electronic Circuits for Medical Applications(2011) Safar, Mohammad A A A; Newcomb, Robert; Electrical Engineering; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)This thesis deals with the theory and design of a hemispherical antenna array circuit that is capable to operate in the intermediate zones. By doing that, this array can be used in Hyperthermia Treatment for Brain Cancer in which the aim is to noninvasively focus the fields at microwave frequencies to the location of the tumor cells in the brain. Another possible application of the array is to offer an alternative means of sustaining Deep Brain Stimulation other than using the traditional (surgical) approach. The new noninvasive technique is accomplished by the use of a hemispherical antenna array placed on the human's head. The array uses a new beamforming technique that achieves 3 dimensional beamforming or focusing of the magnetic field of antennas to desired points in the brain to achieve either cell death by temperature rise (Hyperthermia Application) or to cause brain stimulation and hopefully alleviate the affects of Parkinson's Disease (Deep Brain Stimulation). The main obstacle in this design was that the far field approximation that is usually used when designing antenna arrays does not apply in this case since the hemispherical array is in close proximity to where the magnetic field is desired to be focused. The antenna array problem is approached as a boundary-valued problem with the human head being modeled as a three layered hemisphere. The exact expressions for electromagnetic fields are derived. Health issues such as electric field exposure and specific absorption rate (SAR) are considered. After developing the main antenna and beamforming theory, a neural network is designed to accomplish the beamforming technique used. The radio-frequency (RF) transmitter was designed to transmit the fields at a frequency of 1.8 GHz. The antenna array can also be used as a receiver. The antenna and beamforming theory is presented. A new reception technique is shown which enables the array to receive multiple magnetic field sources from within the hemispherical surface. The receiver is designed to operate at 500 kHz with the RF receiver circuit designed to receive any signal from within the hemispherical surface at a frequency of 500 kHz.