Physical Sciences and Mathematics Commons^™

Modeling The Spatiotemporal Variations Of The Magnetic Field In Active Regions On The Sun Using Deep Neural Networks, Godwill Asare Mensah Mensah

Open Access Theses & Dissertations

Solar active regions are areas on the Sun's surface that have especially strong magnetic fields. Active regions are usually linked to a number of phenomena that can have serious detrimental consequences on technology and, in turn, human life. Examples of these phenomena include solar flares and coronal mass ejections, or CMEs. The precise predictionof solar flares and coronal mass ejections is still an open problem since the fundamental processes underpinning the formation and development of active regions are still not well understood. One key area of research at the intersection of solar physics and artificial intelligence is deriving insights from …

Context-Aware Temporal Embeddings For Text And Video Data, Ahnaf Farhan

Open Access Theses & Dissertations

Recent years have seen an exponential increase in unstructured data, primarily in the form of text, images, and videos. Extracting useful features and trends from large-scale unstructured datasets -- such as news outlets, scientific papers, and videos like security cameras or body cam recordings -- is faced with substantial challenges of volume, scalability, complexity, and semantic understanding. In analyzing trends, comprehending the temporal context is vital for uncovering patterns and narratives that are not apparent from a single video frame or text document. Despite its importance, many existing data mining and machine learning approaches overlook extracting evolutionary contextual features in …

Glacier Segmentation From Remote Sensing Imagery Using Deep Learning, Bibek Aryal

Open Access Theses & Dissertations

Large-scale study of glaciers improves our understanding of global glacier change and is imperative for monitoring the ecological environment, preventing disasters, and studying the effects of global climate change. In recent years, remote sensing imagery has been preferred over riskier and resource-intensive field visits for tracking landscape level changes like glaciers. However, periodic manual labeling of glaciers over a large area is not feasible due to the considerable amount of time it requires while automatic segmentation of glaciers has its own set of challenges. Our work aims to study the challenges associated with segmentation of glaciers from remote sensing imagery …

When Is Deep Learning Better And When Is Shallow Learning Better: Qualitative Analysis, Salvador Robles Herrera, Martine Ceberio, Vladik Kreinovich

Departmental Technical Reports (CS)

In many practical situations, deep neural networks work better than the traditional "shallow" ones, however, in some cases, the shallow neural networks lead to better results. At present, deciding which type of neural networks will work better is mostly done by trial and error. It is therefore desirable to come up with some criterion of when deep learning is better and when shallow is better. In this paper, we argue that this depends on whether the corresponding situation has natural symmetries: if it does, we expect deep learning to work better, otherwise we expect shallow learning to be more effective. …

Hardware For Quantized Mixed-Precision Deep Neural Networks, Andres Rios

Open Access Theses & Dissertations

Recently, there has been a push to perform deep learning (DL) computations on the edge rather than the cloud due to latency, network connectivity, energy consumption, and privacy issues. However, state-of-the-art deep neural networks (DNNs) require vast amounts of computational power, data, and energyâ??resources that are limited on edge devices. This limitation has brought the need to design domain-specific architectures (DSAs) that implement DL-specific hardware optimizations. Traditionally DNNs have run on 32-bit floating-point numbers; however, a body of research has shown that DNNs are surprisingly robust and do not require all 32 bits. Instead, using quantization, networks can run on …

Benchmarking Machine Learning Methods For Molecular Property Prediction, Govinda Bahadur Kc

Open Access Theses & Dissertations

Machine learning (ML) techniques have been widely applied in a variety of areas ranging from pattern recognition, natural language processing, and computer games to self-driving cars, clinical diagnostics, and molecular structure prediction easing day to day life of human beings. Drug discovery is an expensive, complex, and time taking process. Currently, the pharma industry is hoping to leverage machine learning methods in expediting the drug discovery process. Molecular property prediction is one of the most important tasks in drug discovery. While developing a new drug relies on a proper understanding of molecular properties, there has been great interest in the …

Dedicated Hardware For Machine/Deep Learning: Domain Specific Architectures, Angel Izael Solis

Open Access Theses & Dissertations

Artificial intelligence has come a very long way from being a mere spectacle on the silver screen in the 1920s [Hml18]. As artificial intelligence continues to evolve, and we begin to develop more sophisticated Artificial Neural Networks, the need for specialized and more efficient machines (less computational strain while maintaining the same performance results) becomes increasingly evident. Though these “new” techniques, such as Multilayer Perceptron’s, Convolutional Neural Networks and Recurrent Neural Networks, may seem as if they are on the cutting edge of technology, many of these ideas are over 60 years old! However, many of these earlier models, at …

A Novel Set Of Weight Initialization Techniques For Deep Learning Architectures, Diego Aguirre

Open Access Theses & Dissertations

The importance of weight initialization when building a deep learning model is often underappreciated. Even though it is usually seen as a minor detail in the model creation cycle, this process has shown to have a strong impact on the training time of a network and the quality of the resulting model. In fact, the implications of choosing a poor initialization scheme range from leading to the creation of a poorly performing model to preventing optimization techniques (like stochastic gradient descent) from converging.

In this work, we introduce and evaluate a set of novel weight initialization techniques for deep learning …

Deep Learning Models For Scoring Protein-Ligand Interaction Energies, Md Mahmudulla Hassan

Open Access Theses & Dissertations

In recent years, the cheminformatics community has seen an increased success with machine learning-based scoring functions for estimating binding affinities. The prediction of protein-ligand binding affinities is crucial for drug discovery research. Many physics-based scoring functions have been developed over the years. Lately, machine learning approaches are proven to boost the performance of traditional scoring functions. In this study, two scoring functions were developed; one is based on the Convolutional Neural Networks and the other one, called DLSCORE, is based on an ensemble of fully connected neural networks. Both the models were trained on the refined PDBbind (v.2016) dataset using …

Analysing The Effects Of Data Augmentation And Free Parameters For Text Classification With Recurrent Convolutional Neural Networks, Jonathan Quijas

Open Access Theses & Dissertations

Convolutional neural networks have seen much success in computer vision and natural language processing tasks. When training convolutional neural networks for text classification tasks, a common technique is to transform an input sequence of words into a dense matrix of word embeddings, or words represented as dense vectors, using table lookup operations. This enables the inputs to be represented in a way that the well-known convolution/pooling operations can be applied to them in a manner similar to images. These word embeddings may be further incorporated into the neural network itself as a trainable layer to allow fine-tuning, usually leading to …