Computer Sciences Commons^™

Recall Distortion In Neural Network Pruning And The Undecayed Pruning Algorithm, Aidan Good, Jiaqi Lin, Hannah Sieg, Mikey Ferguson, Xin Yu, Shandian Zhe, Jerzy Wieczorek, Thiago Serra

Faculty Conference Papers and Presentations

Pruning techniques have been successfully used in neural networks to trade accuracy for sparsity. However, the impact of network pruning is not uniform: prior work has shown that the recall for underrepresented classes in a dataset may be more negatively affected. In this work, we study such relative distortions in recall by hypothesizing an intensification effect that is inherent to the model. Namely, that pruning makes recall relatively worse for a class with recall below accuracy and, conversely, that it makes recall relatively better for a class with recall above accuracy. In addition, we propose a new pruning algorithm aimed …

Lstm-Sdm: An Integrated Framework Of Lstm Implementation For Sequential Data Modeling[Formula Presented], Hum Nath Bhandari, Binod Rimal, Nawa Raj Pokhrel, Ramchandra Rimal, Keshab R. Dahal

Arts & Sciences Faculty Publications

LSTM-SDM is a python-based integrated computational framework built on the top of Tensorflow/Keras and written in the Jupyter notebook. It provides several object-oriented functionalities for implementing single layer and multilayer LSTM models for sequential data modeling and time series forecasting. Multiple subroutines are blended to create a conducive user-friendly environment that facilitates data exploration and visualization, normalization and input preparation, hyperparameter tuning, performance evaluations, visualization of results, and statistical analysis. We utilized the LSTM-SDM framework in predicting the stock market index and observed impressive results. The framework can be generalized to solve several other real-world time series problems.

Marginal Proportional Hazards Models For Clustered Interval-Censored Data With Time-Dependent Covariates, Kaitlyn Cook, Wenbin Lu, Rui Wang

Statistical and Data Sciences: Faculty Publications

The Botswana Combination Prevention Project was a cluster-randomized HIV prevention trial whose follow-up period coincided with Botswana’s national adoption of a universal test-and-treat strategy for HIV management. Of interest is whether, and to what extent, this change in policy (i) modified the observed preventative effects of the study intervention and (ii) was associated with a reduction in the population-level incidence of HIV in Botswana. To address these questions, we propose a stratified proportional hazards model for clustered intervalcensored data with time-dependent covariates and develop a composite expectation maximization algorithm that facilitates estimation of model parameters without placing parametric assumptions on …

Policy-Based Redactable Signatures, Zachary Kissel

Computer and Data Science Faculty Publications

In this work we make progress towards solving an open problem posed by Bilzhause et. al, to give constructions of redactable signature schemes that allow the signer to limit the possible redactions performed by a third party. A separate, but related notion, called controlled disclosure allows a redactor to limit future redactions. We look at two types of data, sets and linear data (data organized as a sequence). In the case of sets, we limit redactions using a policy modeled by a monotone circuit or any circuit depending on the size of the universe the set is drawn from. In …

An Investigation Of The Reconstruction Capacity Of Stacked Convolutional Autoencoders For Log-Mel-Spectrograms, Anastasia Natsiou, Luca Longo, Seán O'Leary

Conference Papers

In audio processing applications, the generation of expressive sounds based on high-level representations demonstrates a high demand. These representations can be used to manipulate the timbre and influence the synthesis of creative instrumental notes. Modern algorithms, such as neural networks, have inspired the development of expressive synthesizers based on musical instrument timbre compression. Unsupervised deep learning methods can achieve audio compression by training the network to learn a mapping from waveforms or spectrograms to low-dimensional representations. This study investigates the use of stacked convolutional autoencoders for the compression of time-frequency audio representations for a variety of instruments for a single …

Tutorial: Neuro-Symbolic Ai For Mental Healthcare, Kaushik Roy, Usha Lokala, Manas Gaur, Amit Sheth

Publications

Artificial Intelligence (AI) systems for mental healthcare (MHCare) have been ever-growing after realizing the importance of early interventions for patients with chronic mental health (MH) conditions. Social media (SocMedia) emerged as the go-to platform for supporting patients seeking MHCare. The creation of peer-support groups without social stigma has resulted in patients transitioning from clinical settings to SocMedia supported interactions for quick help. Researchers started exploring SocMedia content in search of cues that showcase correlation or causation between different MH conditions to design better interventional strategies. User-level Classification-based AI systems were designed to leverage diverse SocMedia data from various MH conditions, …

Learning To Automate Follow-Up Question Generation Using Process Knowledge For Depression Triage On Reddit Posts, Shrey Gupta, Anmol Agarwal, Manas Gaur, Kaushik Roy, Vignesh Narayanan, Ponnurangam Kumaraguru, Amit Sheth

Publications

Conversational Agents (CAs) powered with deep language models (DLMs) have shown tremendous promise in the domain of mental health. Prominently, the CAs have been used to provide informational or therapeutic services (e.g., cognitive behavioral therapy) to patients. However, the utility of CAs to assist in mental health triaging has not been explored in the existing work as it requires a controlled generation of follow-up questions (FQs), which are often initiated and guided by the mental health professionals (MHPs) in clinical settings. In the context of `depression', our experiments show that DLMs coupled with process knowledge in a mental health questionnaire …

Phishing Detection Using Natural Language Processing And Machine Learning, Apurv Mittal, Dr Daniel Engels, Harsha Kommanapalli, Ravi Sivaraman, Taifur Chowdhury

SMU Data Science Review

Phishing emails are a primary mode of entry for attackers into an organization. A successful phishing attempt leads to unauthorized access to sensitive information and systems. However, automatically identifying phishing emails is often difficult since many phishing emails have composite features such as body text and metadata that are nearly indistinguishable from valid emails. This paper presents a novel machine learning-based framework, the DARTH framework, that characterizes and combines multiple models, with one model for each composite feature, that enables the accurate identification of phishing emails. The framework analyses each composite feature independently utilizing a multi-faceted approach using Natural Language …

Using Natural Language Processing To Increase Modularity And Interpretability Of Automated Essay Evaluation And Student Feedback, Chris Roche, Nathan Deinlein, Darryl Dawkins, Faizan Javed

SMU Data Science Review

For English teachers and students who are dissatisfied with the one-size-fits-all approach of current Automated Essay Scoring (AES) systems, this research uses Natural Language Processing (NLP) techniques that provide a focus on configurability and interpretability. Unlike traditional AES models which are designed to provide an overall score based on pre-trained criteria, this tool allows teachers to tailor feedback based upon specific focus areas. The tool implements a user-interface that serves as a customizable rubric. Students’ essays are inputted into the tool either by the student or by the teacher via the application’s user-interface. Based on the rubric settings, the tool …

Cov-Inception: Covid-19 Detection Tool Using Chest X-Ray, Aswini Thota, Ololade Awodipe, Rashmi Patel

SMU Data Science Review

Since the pandemic started, researchers have been trying to find a way to detect COVID-19 which is a cost-effective, fast, and reliable way to keep the economy viable and running. This research details how chest X-ray radiography can be utilized to detect the infection. This can be for implementation in Airports, Schools, and places of business. Currently, Chest imaging is not a first-line test for COVID-19 due to low diagnostic accuracy and confounding with other viral pneumonia. Different pre-trained algorithms were fine-tuned and applied to the images to train the model and the best model obtained was fine-tuned InceptionV3 model …

Application Of Probabilistic Ranking Systems On Women’S Junior Division Beach Volleyball, Cameron Stewart, Michael Mazel, Bivin Sadler

SMU Data Science Review

Women’s beach volleyball is one of the fastest growing collegiate sports today. The increase in popularity has come with an increase in valuable scholarship opportunities across the country. With thousands of athletes to sort through, college scouts depend on websites that aggregate tournament results and rank players nationally. This project partnered with the company Volleyball Life, who is the current market leader in the ranking space of junior beach volleyball players. Utilizing the tournament information provided by Volleyball Life, this study explored replacements to the current ranking systems, which are designed to aggregate player points from recent tournament placements. Three …

“Be A Pattern For The World”: The Development Of A Dark Patterns Detection Tool To Prevent Online User Loss, Jordan Donnelly, Alan Dowley, Yunpeng Liu, Yufei Su, Quanwei Sun, Lan Zeng, Andrea Curley, Damian Gordon, Paul Kelly, Dympna O'Sullivan, Anna Becevel

Articles

Dark Patterns are designed to trick users into sharing more information or spending more money than they had intended to do, by configuring online interactions to confuse or add pressure to the users. They are highly varied in their form, and are therefore difficult to classify and detect. Therefore, this research is designed to develop a framework for the automated detection of potential instances of web-based dark patterns, and from there to develop a software tool that will provide a highly useful defensive tool that helps detect and highlight these patterns.

Self-Supervised Learning For Invariant Representations From Multi-Spectral And Sar Images, Pallavi Jain, Bianca Schoen Phelan, Robert J. Ross

Articles

Self-Supervised learning (SSL) has become the new state of the art in several domain classification and segmentation tasks. One popular category of SSL are distillation networks such as Bootstrap Your Own Latent (BYOL). This work proposes RS-BYOL, which builds on BYOL in the remote sensing (RS) domain where data are non-trivially different from natural RGB images. Since multi-spectral (MS) and synthetic aperture radar (SAR) sensors provide varied spectral and spatial resolution information, we utilise them as an implicit augmentation to learn invariant feature embeddings. In order to learn RS based invariant features with SSL, we trained RS-BYOL in two ways, …

Implementing Github Actions Continuous Integration To Reduce Error Rates In Ecological Data Collection, Albert Y. Kim, Valentine Herrmann, Ross Barreto, Brianna Calkins, Erika Gonzalez-Akre, Daniel J. Johnson, Jennifer A. Jordan, Lukas Magee, Ian R. Mcgregor, Nicolle Montero, Karl Novak, Teagan Rogers, Jessica Shue, Kristina J. Anderson-Teixeira

Statistical and Data Sciences: Faculty Publications

Accurate field data are essential to understanding ecological systems and forecasting their responses to global change. Yet, data collection errors are common, and data analysis often lags far enough behind its collection that many errors can no longer be corrected, nor can anomalous observations be revisited. Needed is a system in which data quality assurance and control (QA/QC), along with the production of basic data summaries, can be automated immediately following data collection.

Here, we implement and test a system to satisfy these needs. For two annual tree mortality censuses and a dendrometer band survey at two forest research sites, …

Data-Centric Machine Learning For Speech And Audio, Ali Raza Syed

Dissertations, Theses, and Capstone Projects

There is growing recognition of the importance of data-centric methods for building machine learning systems. Data-centric methods assume a fixed model and iterate over the data to improve system performance. This is in contrast to traditional model-centric approaches, which assume a fixed dataset and iterate over models for the same ends. Data-centric machine learning is driven by the observation that, beyond the size of the training data, model performance depends on factors such as the quality of the annotations, and whether the data are representative of conditions in which models will be deployed. This is particularly of interest in the …

The Interaction Of Different Primary Producers And Physical And Chemical Dynamics Of An Urban Shallow Lake, Majid Sahin

Dissertations, Theses, and Capstone Projects

An artificial urban shallow lake, Prospect Park Lake (PPL), is situated on a terminal moraine in Brooklyn New York, and supplied with municipal water treated with ortho-phosphates. The constant input of the phosphate nutrient is the primary source of eutrophication in the lake. The numerous pools along the water course houses various aquatic phototrophs, which influence the water quality and the state of the system, driving conditions into favoring the survival of their species. In the first half of the dissertation, the focus of the project is on analyzing how the different primary producers in different regions of PPL affect …

Finite Gaussian Neurons: Defending Against Adversarial Attacks By Making Neural Networks Say "I Don’T Know", Felix Grezes

Dissertations, Theses, and Capstone Projects

In this work, I introduce the Finite Gaussian Neuron (FGN), a novel neuron architecture for artificial neural networks aimed at protecting against adversarial attacks.
Since 2014, artificial neural networks have been known to be vulnerable to adversarial attacks, which can fool the network into producing wrong or nonsensical outputs by making humanly imperceptible alterations to inputs. While defenses against adversarial attacks have been proposed, they usually involve retraining a new neural network from scratch, a costly task.

My works aims to:
- easily convert existing models to Finite Gaussian Neuron architecture,
- while preserving the existing model's behavior on real …

Efficient And Scalable Triangle Centrality Algorithms In The Arkouda Framework, Joseph Thomas Patchett

Theses

Graph data structures provide a unique challenge for both analysis and algorithm development. These data structures are irregular in that memory accesses are not known a priori and accesses to these structures tend to lack locality.

Despite these challenges, graph data structures are a natural way to represent relationships between entities and to exhibit unique features about these relationships. The network created from these relationships can create unique local structures that can describe the behavior between members of these structures. Graphs can be analyzed in a number of different ways including at a high level in community detection and at …

Secure Human Action Recognition By Encrypted Neural Network Inference, Miran Kim, Xiaoqian Jiang, Kristin Lauter, Elkhan Ismayilzada, Shayan Shams

Faculty, Staff and Student Publications

Advanced computer vision technology can provide near real-time home monitoring to support "aging in place" by detecting falls and symptoms related to seizures and stroke. Affordable webcams, together with cloud computing services (to run machine learning algorithms), can potentially bring significant social benefits. However, it has not been deployed in practice because of privacy concerns. In this paper, we propose a strategy that uses homomorphic encryption to resolve this dilemma, which guarantees information confidentiality while retaining action detection. Our protocol for secure inference can distinguish falls from activities of daily living with 86.21% sensitivity and 99.14% specificity, with an average …

Mathematical Models Yield Insights Into Cnns: Applications In Natural Image Restoration And Population Genetics, Ryan Cecil

Electronic Theses and Dissertations

Due to a rise in computational power, machine learning (ML) methods have become the state-of-the-art in a variety of fields. Known to be black-box approaches, however, these methods are oftentimes not well understood. In this work, we utilize our understanding of model-based approaches to derive insights into Convolutional Neural Networks (CNNs). In the field of Natural Image Restoration, we focus on the image denoising problem. Recent work have demonstrated the potential of mathematically motivated CNN architectures that learn both `geometric' and nonlinear higher order features and corresponding regularizers. We extend this work by showing that not only can geometric features …

Automated Identification Of Astronauts On Board The International Space Station: A Case Study In Space Archaeology, Rao Hamza Ali, Amir Kanan Kashefi, Alice C. Gorman, Justin St. P. Walsh, Erik J. Linstead

Art Faculty Articles and Research

We develop and apply a deep learning-based computer vision pipeline to automatically identify crew members in archival photographic imagery taken on-board the International Space Station. Our approach is able to quickly tag thousands of images from public and private photo repositories without human supervision with high degrees of accuracy, including photographs where crew faces are partially obscured. Using the results of our pipeline, we carry out a large-scale network analysis of the crew, using the imagery data to provide novel insights into the social interactions among crew during their missions.

Evaluating The Variable Stride Algorithm In The Identification Of Diabetic Retinopathy, Ying Zheng, Brian Danaher, Matthew Brown

Beyond: Undergraduate Research Journal

An experiment was performed to investigate a modified pooling method for use in convolutional neural networks for image recognition. This algorithm–Variable Stride–allows the user to segment an image and change the amount of subsampling in each region. This control allows for the user to maintain a higher amount of data retention in more important regions of the image, while more aggressively subsampling the less important regions to increase training speed. Three Variable Stride methods were compared to the preexisting pooling algorithms, Maximum Pool and Average Pool, in three different network configurations tasked with classifying Diabetic Retinopathy images between its early …

Perturbation Modeling For Molecular Design Of Protein Tyrosine Kinase Inhibitors Using Unsupervised Machine Learning, Keerthi Krishnan

Computational and Data Sciences (MS) Theses

The field of computational drug discovery and development has grown, with the aid of new computational tools for novel molecule discovery. In specific, generative deep learning models have excelled as tools to aid in navigating the large space of known molecules and in the creation of new molecules. These models are fed various representations of molecules as inputs and learn to perform a variety of things, such as the optimization of these molecules towards a targeted property. Ultimately, these generative learning models allow us to build bridges between chemical and continuous spaces to understand the compromise between invoking small incremental …

Data Collection And Machine Learning Methods For Automated Pedestrian Facility Detection And Mensuration, Joseph Bailey Luttrell Iv

Dissertations

Large-scale collection of pedestrian facility (crosswalks, sidewalks, etc.) presence data is vital to the success of efforts to improve pedestrian facility management, safety analysis, and road network planning. However, this kind of data is typically not available on a large scale due to the high labor and time costs that are the result of relying on manual data collection methods. Therefore, methods for automating this process using techniques such as machine learning are currently being explored by researchers. In our work, we mainly focus on machine learning methods for the detection of crosswalks and sidewalks from both aerial and street-view …

Parallel Algorithms For Scalable Graph Mining: Applications On Big Data And Machine Learning, Naw Safrin Sattar

LSU New Orleans Theses and Dissertations

Parallel computing plays a crucial role in processing large-scale graph data. Complex network analysis is an exciting area of research for many applications in different scientific domains e.g., sociology, biology, online media, recommendation systems and many more. Graph mining is an area of interest with diverse problems from different domains of our daily life. Due to the advancement of data and computing technologies, graph data is growing at an enormous rate, for example, the number of links in social networks is growing every millisecond. Machine/Deep learning plays a significant role for technological accomplishments to work with big data in modern …

Ocean Wave Prediction And Characterization For Intelligent Maritime Transportation, Pujan Pokhrel

LSU New Orleans Theses and Dissertations

The national Earth System Prediction (ESPC) initiative aims to develop the predictions
for the next generation predictions of atmosphere, ocean, and sea-ice interactions in the scale of days to decades. This dissertation seeks to demonstrate the methods we can use to improve the ESPC models, especially the ocean prediction model. In the application side of the weather forecasts, this dissertation explores imitation learning with constraints to solve combinatorial optimization problems, focusing on the weather routing of surface vessels. Prediction of ocean waves is essential for various purposes, including vessel routing, ocean energy harvesting, agriculture, etc. Since the machine learning approaches …

Solving The Challenges Of Concept Drift In Data Stream Classification., Hanqing Hu

Electronic Theses and Dissertations

The rise of network connected devices and applications leads to a significant increase in the volume of data that are continuously generated overtime time, called data streams. In real world applications, storing the entirety of a data stream for analyzing later is often not practical, due to the data stream’s potentially infinite volume. Data stream mining techniques and frameworks are therefore created to analyze streaming data as they arrive. However, compared to traditional data mining techniques, challenges unique to data stream mining also emerge, due to the high arrival rate of data streams and their dynamic nature. In this dissertation, …

Computer Aided Diagnosis System For Breast Cancer Using Deep Learning., Asma Baccouche

Electronic Theses and Dissertations

The recent rise of big data technology surrounding the electronic systems and developed toolkits gave birth to new promises for Artificial Intelligence (AI). With the continuous use of data-centric systems and machines in our lives, such as social media, surveys, emails, reports, etc., there is no doubt that data has gained the center of attention by scientists and motivated them to provide more decision-making and operational support systems across multiple domains. With the recent breakthroughs in artificial intelligence, the use of machine learning and deep learning models have achieved remarkable advances in computer vision, ecommerce, cybersecurity, and healthcare. Particularly, numerous …

Unsupervised Contrastive Representation Learning For Knowledge Distillation And Clustering, Fei Ding

All Dissertations

Unsupervised contrastive learning has emerged as an important training strategy to learn representation by pulling positive samples closer and pushing negative samples apart in low-dimensional latent space. Usually, positive samples are the augmented versions of the same input and negative samples are from different inputs. Once the low-dimensional representations are learned, further analysis, such as clustering, and classification can be performed using the representations. Currently, there are two challenges in this framework. First, the empirical studies reveal that even though contrastive learning methods show great progress in representation learning on large model training, they do not work well for small …

Computational Models To Detect Radiation In Urban Environments: An Application Of Signal Processing Techniques And Neural Networks To Radiation Data Analysis, Jose Nicolas Gachancipa

Beyond: Undergraduate Research Journal

Radioactive sources, such as uranium-235, are nuclides that emit ionizing radiation, and which can be used to build nuclear weapons. In public areas, the presence of a radioactive nuclide can present a risk to the population, and therefore, it is imperative that threats are identified by radiological search and response teams in a timely and effective manner. In urban environments, such as densely populated cities, radioactive sources may be more difficult to detect, since background radiation produced by surrounding objects and structures (e.g., buildings, cars) can hinder the effective detection of unnatural radioactive material. This article presents a computational model …