Physical Sciences and Mathematics Commons^™

Optimizing Collective Communication For Scalable Scientific Computing And Deep Learning, Jiali Li

Doctoral Dissertations

In the realm of distributed computing, collective operations involve coordinated communication and synchronization among multiple processing units, enabling efficient data exchange and collaboration. Scientific applications, such as simulations, computational fluid dynamics, and scalable deep learning, require complex computations that can be parallelized across multiple nodes in a distributed system. These applications often involve data-dependent communication patterns, where collective operations are critical for achieving high performance in data exchange. Optimizing collective operations for scientific applications and deep learning involves improving the algorithms, communication patterns, and data distribution strategies to minimize communication overhead and maximize computational efficiency.

Within the context of this …

Insights Into The Application Of Deep Reinforcement Learning In Healthcare And Materials Science, Benjamin R. Smith

Doctoral Dissertations

Reinforcement learning (RL) is a type of machine learning designed to optimize sequential decision-making. While controlled environments have served as a foundation for RL research, due to the growth in data volumes and deep learning methods, it is now increasingly being applied to real-world problems. In our work, we explore and attempt to overcome challenges that occur when applying RL to solve problems in healthcare and materials science.

First, we explore how issues in bias and data completeness affect healthcare applications of RL. To understand how bias has already been considered in this area, we survey the literature for existing …

Learning From Sequential User Data: Models And Sample-Efficient Algorithms, Aritra Ghosh

Doctoral Dissertations

Recent advances in deep learning have made learning representation from ever-growing datasets possible in the domain of vision, natural language processing (NLP), and robotics, among others. However, deep networks are notoriously data-hungry; for example, training language models with attention mechanisms sometimes requires trillions of parameters and tokens. In contrast, we can often access a limited number of samples in many tasks. It is crucial to learn models from these `limited' datasets. Learning with limited datasets can take several forms. In this thesis, we study how to select data samples sequentially such that downstream task performance is maximized. Moreover, we study …

On Interpreting Eddy Covariance In Small Area Agricultural Situations With Contrasting Site Management., Joel Oetting

Doctoral Dissertations

This dissertation examined the carbon sequestration potential of a low C:N soil amendment and its incorporation into the soil over a rolling agricultural field. A segmented planar fit was developed to assess and correct the systematic errors the topography introduces on the carbon dioxide fluxes. The carbon dioxide fluxes were then be partitioned into gross primary productivity and soil respiration to understand the influence of the contrasting management practices, using flux variance partitioning. Concomitant with the partitioning, high resolution temporal and spatial scale remote sensing images were interpolated and standardized to conduct hypothesis testing for treatment effects.

Social Media Analytics With Applications In Disaster Management And Covid-19 Events, Md Yasin Kabir

Doctoral Dissertations

"Social media such as Twitter offers a tremendous amount of data throughout an event or a disastrous situation. Leveraging social media data during a disaster is beneficial for effective and efficient disaster management. Information extraction, trend identification, and determining public reactions might help in the future disaster or even avert such an event. However, during a disaster situation, a robust system is required that can be deployed faster and process relevant information with satisfactory performance in real-time. This work outlines the research contributions toward developing such an effective system for disaster management, where it is paramount to develop automated machine-enabled …

Remote Sensing Of High Latitude Rivers: Approaches, Insights, And Future Ramifications, Merritt E. Harlan

Doctoral Dissertations

High latitude rivers across the pan-Arctic domain are changing due to changes in climate and positive Arctic feedback loops. Understanding and contextualizing these changes is challenging due to a lack of data and methods for estimating and modeling river discharge, and mapping rivers. Remote sensing, and the availability of satellite imagery can provide ways to overcome these challenges. Through combining various forms of fieldwork, modeling, deep learning, and remote sensing, we contribute methodologies and knowledge to three key challenges associated with better understanding high latitude rivers. In the first chapter, we combine field data that can be rapidly deployed with …

Deep Learning-Based Surrogate Models For Post-Earthquake Damage Assessment, Xinzhe Yuan

Doctoral Dissertations

"Seismic damage assessment is a critical step to enhance community resilience in the wake of an earthquake. This study aims to develop deep learning-based surrogate models for widely used fragility curves to achieve more accurate and rapid assessment in practice. These surrogate models are based on artificial neural networks trained from the labelled ground motions whose resulting damage classes on targeted structures are determined by nonlinear time history analyses. The development of various surrogate models is progressed in four phases. In Phase I, the multilayer perceptron (MLP) is used to develop multivariate seismic classifiers with up to 50 hand-crafted intensity …

Auto-Curation Of Large Evolving Image Datasets, Sara Mousavicheshmehkaboodi

Doctoral Dissertations

Large image collections are becoming common in many fields and offer tantalizing opportunities to transform how research, work, and education are conducted if the information and associated insights could be extracted from them. However, major obstacles to this vision exist. First, image datasets with associated metadata contain errors and need to be cleaned and organized to be easily explored and utilized. Second, such collections typically lack the necessary context or may have missing attributes that need to be recovered. Third, such datasets are domain-specific and require human expert involvement to make the right interpretation of the image content. Fourth, the …

Deep Learning Models For Irregularly Sampled And Incomplete Time Series, Satya Narayan Shukla

Doctoral Dissertations

Irregularly sampled time series data arise naturally in many application domains including biology, ecology, climate science, astronomy, geology, finance, and health. Such data present fundamental challenges to many classical models from machine learning and statistics. The first challenge with modeling such data is the presence of variable time gaps between the observation time points. The second challenge is that the dimensionality of the inputs can be different for different data cases. This occurs naturally due to the fact that different data cases are likely to include different numbers of observations. The third challenge is that different irregularly sampled instances have …

Utilizing Graph Structure For Machine Learning, Stefan Dernbach

Doctoral Dissertations

The information age has led to an explosion in the size and availability of data. This data often exhibits graph-structure that is either explicitly defined, as in the web of a social network, or is implicitly defined and can be determined by measuring similarity between objects. Utilizing this graph-structure allows for the design of machine learning algorithms that reflect not only the attributes of individual objects but their relationships to every other object in the domain as well. This thesis investigates three machine learning problems and proposes novel methods that leverage the graph-structure inherent in the tasks. Quantum walk neural …

Neural Methods For Answer Passage Retrieval Over Sparse Collections, Daniel Cohen

Doctoral Dissertations

Recent advances in machine learning have allowed information retrieval (IR) techniques to advance beyond the stage of handcrafting domain specific features. Specifically, deep neural models incorporate varying levels of features to learn whether a document answers the information need of a query. However, these neural models rely on a large number of parameters to successfully learn a relation between a query and a relevant document.

This reliance on a large number of parameters, combined with the current methods of optimization relying on small updates necessitates numerous samples to allow the neural model to converge on an effective relevance function. This …

Improving Visual Recognition With Unlabeled Data, Aruni Roy Chowdhury

Doctoral Dissertations

The success of deep neural networks has resulted in computer vision systems that obtain high accuracy on a wide variety of tasks such as image classification, object detection, semantic segmentation, etc. However, most state-of-the-art vision systems are dependent upon large amounts of labeled training data, which is not a scalable solution in the long run. This work focuses on improving existing models for visual object recognition and detection without being dependent on such large-scale human-annotated data. We first show how large numbers of hard examples (cases where an existing model makes a mistake) can be obtained automatically from unlabeled video …

Higher-Order Representations For Visual Recognition, Tsung-Yu Lin

Doctoral Dissertations

In this thesis, we present a simple and effective architecture called Bilinear Convolutional Neural Networks (B-CNNs). These networks represent an image as a pooled outer product of features derived from two CNNs and capture localized feature interactions in a translationally invariant manner. B-CNNs generalize classical orderless texture-based image models such as bag-of-visual-words and Fisher vector representations. However, unlike prior work, they can be trained in an end-to-end manner. In the experiments, we demonstrate that these representations generalize well to novel domains by fine-tuning and achieve excellent results on fine-grained, texture and scene recognition tasks. The visualization of fine-tuned convolutional filters …

Improving Face Clustering In Videos, Souyoung Jin

Doctoral Dissertations

Human faces represent not only a challenging recognition problem for computer vision, but are also an important source of information about identity, intent, and state of mind. These properties make the analysis of faces important not just as algorithmic challenges, but as a gateway to developing computer vision methods that can better follow the intent and goals of human beings. In this thesis, we are interested in face clustering in videos. Given a raw video, with no caption or annotation, we want to group all detected faces by their identity. We address three problems in the area of face clustering …

Deep Learning For Digitized Histology Image Analysis, Sudhir Sornapudi

Doctoral Dissertations

“Cervical cancer is the fourth most frequent cancer that affects women worldwide. Assessment of cervical intraepithelial neoplasia (CIN) through histopathology remains as the standard for absolute determination of cancer. The examination of tissue samples under a microscope requires considerable time and effort from expert pathologists. There is a need to design an automated tool to assist pathologists for digitized histology slide analysis. Pre-cervical cancer is generally determined by examining the CIN which is the growth of atypical cells from the basement membrane (bottom) to the top of the epithelium. It has four grades, including: Normal, CIN1, CIN2, and CIN3. In …

Neural Models For Information Retrieval Without Labeled Data, Hamed Zamani

Doctoral Dissertations

Recent developments of machine learning models, and in particular deep neural networks, have yielded significant improvements on several computer vision, natural language processing, and speech recognition tasks. Progress with information retrieval (IR) tasks has been slower, however, due to the lack of large-scale training data as well as neural network models specifically designed for effective information retrieval. In this dissertation, we address these two issues by introducing task-specific neural network architectures for a set of IR tasks and proposing novel unsupervised or \emph{weakly supervised} solutions for training the models. The proposed learning solutions do not require labeled training data. Instead, …

Efficient Self-Supervised Deep Sensorimotor Learning In Robotics, Takeshi Takahashi

Doctoral Dissertations

Deep learning has been successful in a variety of applications, such as object recognition, video games, and machine translation. Deep neural networks can automatically learn important features given large training datasets. However, the success of deep learning in robotic systems in the real world is still limited mainly because obtaining large datasets and labeling are costly. As a result, much of the successful work in deep learning has been limited to domains where large datasets are readily available or easily collected. To address this issue, I propose a framework for acquiring re-usable skills efficiently combining intrinsic motivation and the control …

Machine Learning Models For Efficient And Robust Natural Language Processing, Emma Strubell

Doctoral Dissertations

Natural language processing (NLP) has come of age. For example, semantic role labeling (SRL), which automatically annotates sentences with a labeled graph representing who did what to whom, has in the past ten years seen nearly 40% reduction in error, bringing it to useful accuracy. As a result, a myriad of practitioners now want to deploy NLP systems on billions of documents across many domains. However, state-of-the-art NLP systems are typically not optimized for cross-domain robustness nor computational efficiency. In this dissertation I develop machine learning methods to facilitate fast and robust inference across many common NLP tasks. First, …

Applications Of Machine Learning In Nuclear Imaging And Radiation Detection, Shaikat Mahmood Galib

Doctoral Dissertations

"The main focus of this work is to use machine learning and data mining techniques to address some challenging problems that arise from nuclear data. Specifically, two problem areas are discussed: nuclear imaging and radiation detection. The techniques to approach these problems are primarily based on a variant of Artificial Neural Network (ANN) called Convolutional Neural Network (CNN), which is one of the most popular forms of 'deep learning' technique.

The first problem is about interpreting and analyzing 3D medical radiation images automatically. A method is developed to identify and quantify deformable image registration (DIR) errors from lung CT scans …

Deep-Learned Generative Representations Of 3d Shape Families, Haibin Huang

Doctoral Dissertations

Digital representations of 3D shapes are becoming increasingly useful in several emerging applications, such as 3D printing, virtual reality and augmented reality. However, traditional modeling softwares require users to have extensive modeling experience, artistic skills and training to handle their complex interfaces and perform the necessary low-level geometric manipulation commands. Thus, there is an emerging need for computer algorithms that help novice and casual users to quickly and easily generate 3D content. In this work, I will present deep learning algorithms that are capable of automatically inferring parametric representations of shape families, which can be used to generate new 3D …

Deep Energy-Based Models For Structured Prediction, David Belanger

Doctoral Dissertations

We introduce structured prediction energy networks (SPENs), a flexible frame- work for structured prediction. A deep architecture is used to define an energy func- tion over candidate outputs and predictions are produced by gradient-based energy minimization. This deep energy captures dependencies between labels that would lead to intractable graphical models, and allows us to automatically discover discrim- inative features of the structured output. Furthermore, practitioners can explore a wide variety of energy function architectures without having to hand-design predic- tion and learning methods for each model. This is because all of our prediction and learning methods interact with the energy …

Conditional Computation In Deep And Recurrent Neural Networks, Andrew Scott Davis

Doctoral Dissertations

Recently, deep learning models such as convolutional and recurrent neural networks have displaced state-of-the-art techniques in a variety of application domains. While the computationally heavy process of training is usually conducted on powerful graphics processing units (GPUs) distributed in large computing clusters, the resulting models can still be somewhat heavy, making deployment in resource- constrained environments potentially problematic. In this work, we build upon the idea of conditional computation, where the model is given the capability to learn how to avoid computing parts of the graph. This allows for models where the number of parameters (and in a sense, the …

Unsupervised Joint Alignment, Clustering And Feature Learning, Mohamed Marwan Mattar

Doctoral Dissertations

Joint alignment is the process of transforming instances in a data set to make them more similar based on a pre-defined measure of joint similarity. This process has great utility and applicability in many scientific disciplines including radiology, psychology, linguistics, vision, and biology. Most alignment algorithms suffer from two shortcomings. First, they typically fail when presented with complex data sets arising from multiple modalities such as a data set of normal and abnormal heart signals. Second, they require hand-picking appropriate feature representations for each data set, which may be time-consuming and ineffective, or outside the domain of expertise for practitioners. …

Incorporating Boltzmann Machine Priors For Semantic Labeling In Images And Videos, Andrew Kae

Doctoral Dissertations

Semantic labeling is the task of assigning category labels to regions in an image. For example, a scene may consist of regions corresponding to categories such as sky, water, and ground, or parts of a face such as eyes, nose, and mouth. Semantic labeling is an important mid-level vision task for grouping and organizing image regions into coherent parts. Labeling these regions allows us to better understand the scene itself as well as properties of the objects in the scene, such as their parts, location, and interaction within the scene. Typical approaches for this task include the conditional random field …

Online Multi-Stage Deep Architectures For Feature Extraction And Object Recognition, Derek Christopher Rose

Doctoral Dissertations

Multi-stage visual architectures have recently found success in achieving high classification accuracies over image datasets with large variations in pose, lighting, and scale. Inspired by techniques currently at the forefront of deep learning, such architectures are typically composed of one or more layers of preprocessing, feature encoding, and pooling to extract features from raw images. Training these components traditionally relies on large sets of patches that are extracted from a potentially large image dataset. In this context, high-dimensional feature space representations are often helpful for obtaining the best classification performances and providing a higher degree of invariance to object transformations. …