Physical Sciences and Mathematics Commons^™

Adaptable And Trustworthy Machine Learning For Human Activity Recognition From Bioelectric Signals, Morgan S. Stuart

Theses and Dissertations

Enabling machines to learn measures of human activity from bioelectric signals has many applications in human-machine interaction and healthcare. However, labeled activity recognition datasets are costly to collect and highly varied, which challenges machine learning techniques that rely on large datasets. Furthermore, activity recognition in practice needs to account for user trust - models are motivated to enable interpretability, usability, and information privacy. The objective of this dissertation is to improve adaptability and trustworthiness of machine learning models for human activity recognition from bioelectric signals. We improve adaptability by developing pretraining techniques that initialize models for later specialization to unseen …

Tornado Outbreak False Alarm Probabilistic Forecasts With Machine Learning, Kirsten Reed Snodgrass

Theses and Dissertations

Tornadic outbreaks occur annually, causing fatalities and millions of dollars in damage. By improving forecasts, the public can be better equipped to act prior to an event. False alarms (FAs) can hinder the public’s ability (or willingness) to act. As such, a probabilistic FA forecasting scheme would be beneficial to improving public response to outbreaks.

Here, a machine learning approach is employed to predict FA likelihood from Storm Prediction Center (SPC) tornado outbreak forecasts. A database of hit and FA outbreak forecasts spanning 2010 – 2020 was developed using historical SPC convective outlooks and the SPC Storm Reports database. Weather …

Zero And Few-Shot Concept Learning With Pre-Trained Embeddings, Jamison M. Moody

Theses and Dissertations

Neural networks typically struggle with reasoning tasks on out of domain data, something that humans can more easily adapt to. Humans come with prior knowledge of concepts and can segment their environment into building blocks (such as objects) that allow them to reason effectively in unfamiliar situations. Using this intuition, we train a network that utilizes fixed embeddings from the CLIP (Contrastive Language--Image Pre-training) model to do a simple task that the original CLIP model struggles with. The network learns concepts (such as "collide" and "avoid") in a supervised source domain in such a way that the network can adapt …

A Language-Model-Based Chatbot That Considers The User's Personality Profile And Emotions To Support Caregivers Of People With Dementia, Yeganeh Nasiri

Theses and Dissertations

Chatbots are programs that mimic human conversation using Artificial Intelligence (AI). Recent advances in natural language pro- cessing pave the way for chatbots to generate more human-like responses. Therefore, chatbots are finding more complex tasks to perform, such as emotional support which requires both understanding emotions and the ability to properly respond to them. This work presents a chatbot capable of identifying the user's personality and creating responses based on that. During this process, emotion detection is being used to detect and react to users' emotions. The chatbot uses a dynamic knowledge graph to save information as the conversation goes …

Hardware Isolation Approach To Securely Use Untrusted Gpus In Cloud Environments For Machine Learning, Lucas D. Hall

Theses and Dissertations

Machine Learning (ML) is now a primary method for getting useful information out of the immense volumes of data being generated and stored in society today. Useful data is a commodity for training ML models and those that need data for training are often not the owners of the data leading to a desire to use cloud-based services. Deep learning algorithms are best suited to run on a graphical processing unit (GPU) which presents a specific problem since the GPU is not a secure or trusted piece of hardware in the cloud computing environment.

In this paper, we will analyze …

Performance Of Heterogeneous Multi-Agent Systems With Applications In Combined Arms, Robert J. Wilson

Theses and Dissertations

Multi-agent systems show great potential for solving problems in complex and dynamic domains. Such systems comprise multiple individual entities called agents. Agents possessing the same behavior or physical form are called homogeneous while agents which differ in these respects are termed heterogeneous. The overall behavior of the system emerges from the many interactions of its component agents. Most multi-agent systems research to date focuses on systems of homogeneous agents, but recent work suggests that heterogeneous agents may improve system performance in certain tasks. This research examines the impact of heterogeneity on multi-agent system effectiveness and investigates the application of multi-agent …

Team Air Combat Using Model-Based Reinforcement Learning, David A. Mottice

Theses and Dissertations

We formulate the first generalized air combat maneuvering problem (ACMP), called the MvN ACMP, wherein M friendly AUCAVs engage against N enemy AUCAVs, developing a Markov decision process (MDP) model to control the team of M Blue AUCAVs. The MDP model leverages a 5-degree-of-freedom aircraft state transition model and formulates a directed energy weapon capability. Instead, a model-based reinforcement learning approach is adopted wherein an approximate policy iteration algorithmic strategy is implemented to attain high-quality approximate policies relative to a high performing benchmark policy. The ADP algorithm utilizes a multi-layer neural network for the value function approximation regression mechanism. One-versus-one …

Evaluating Semantic Matching Techniques For Technical Documents, Rain F. Dartt

Theses and Dissertations

Machine learning models that employ NLP techniques have become more widely accessible, making them an attractive solution for text and document classification tasks traditionally accomplished by humans. Two such use cases are matching the specialized experience required for a job to statements in applicant resumes, and finding and labelling clauses in legal contracts The AFMC has an immediate need for solutions to civilian hiring. However, there is currently no truth data to validate against. A similar task is contract understanding for which there is the CUAD, a recently published repository of 510 contracts manually labelled by legal experts. The presented …

Smoothing Of Convolutional Neural Network Classifications, Glen R. Drumm

Theses and Dissertations

Smoothing convolutional neural networks is investigated. When intermittent and random false predictions happen, a technique of average smoothing is applied to smooth out the incorrect predictions. While a simple problem environment shows proof of concept, obstacles remain for applying such a technique to a more operationally complex problem.

Obsolescence: Evaluating An Educational Serious Game On Artificial Intelligence Impacts To Military Strategic Goals, Timothy C. Kokotajlo

Theses and Dissertations

Artificial Intelligence (AI) threatens to bring significant disruption to all aspects of military operations. This research develops a Serious Game (SG) and assessment methodology to provide education on the mindsets required for engaging with disruptive AI technologies. The game, Obsolescence, teaches strategic-level concepts recommended to the Department of Defense (DoD) from a compilation of reports on the current and future state of AI and warfighting. The methodology for assessing the educational value of Obsolescence addresses common challenges such as subjective reporting, control groups, population sizes, and measuring abstract or high levels of learning. The games proposed educational value is tested …

Characterizing Convolutional Neural Network Early-Learning And Accelerating Non-Adaptive, First-Order Methods With Localized Lagrangian Restricted Memory Level Bundling, Benjamin O. Morris

Theses and Dissertations

This dissertation studies the underlying optimization problem encountered during the early-learning stages of convolutional neural networks and introduces a training algorithm competitive with existing state-of-the-art methods. First, a Design of Experiments method is introduced to systematically measure empirical second-order Lipschitz upper bound and region size estimates for local regions of convolutional neural network loss surfaces experienced during the early-learning stages. This method demonstrates that architecture choices can significantly impact the local loss surfaces traversed during training. Next, a Design of Experiments method is used to study the effects convolutional neural network architecture hyperparameters have on different optimization routines' abilities to …

Predicting Human Behavior In Repeated Games With Attitude Vectors, Brian L. James

Theses and Dissertations

As Artificial Intelligence systems are used by human users at an increasing frequency, the need for such systems to understand and predict human behavior likewise increases. In my work, I have considered how to predict human behavior in repeated games. These repeated games can be applied as a foundation to many situations where a person may interact with an AI, In an attempt to create such a foundation, I have built a system using Attitude Vectors used in automata to predict actions based on prior actions and communications. These Attitude Vector Automata (AVA) can transform information from actions in one …

Applied Machine Learning In Extrusion-Based Bioprinting, Shuyu Tian

Theses and Dissertations

Optimization of extrusion-based bioprinting (EBB) parameters have been systematically conducted through experimentation. However, the process is time and resource-intensive and not easily translatable across different laboratories. A machine learning (ML) approach to EBB parameter optimization can accelerate this process for laboratories across the field through training using data collected from published literature. In this work, regression-based and classification-based ML models were investigated for their abilities to predict printing outcomes of cell viability and filament diameter for cell-containing alginate and gelatin composite hydrogels. Regression-based models were investigated for their ability to predict suitable extrusion pressure given desired cell viability when keeping …

Neural Network Development In An Artificial Intelligence Gomoku Program, David Garcia

Theses and Dissertations

The game of Gomoku, also called Five in a Row, is an abstract strategy board game. The Gomoku program is constructed upon an algebraic monomial theory to aid values for each possible move and estimate chances for the artificial intelligence program to accomplish a winning path for each move and rounds. With the utilization of the monomial theory, winning configurations are successfully converted into monomials of variables which are represented on board positions. In the artificial intelligence program, an arduous task is how to perform the present configuration of the Gomoku game along with the past moves of the two …

Artificial Intelligence In A Main Warehouse In Panasonic: Los Indios, Texas, Edison Antonio Trejo Hernandez

Theses and Dissertations

The Panasonic Company warehouse is located in Los Indios Texas. The warehouse presents the limitation of the great distances between its headquarters and the Main Warehouse for supplying the branches and main customers, which requires a considerable amount of time to maintain effective communication in the inventory area. In addition, during an online review, it can be confirmed that the website is disabled, contradicting its corporate policy.

The structure of the thesis proposal is arranged in four chapters from the Introduction, Statement of the Problem and Purposes; Previous Studies and Definition of the literature; the Research Methodology and the resources …

Monte Carlo Tree Search Applied To A Modified Pursuit/Evasion Scotland Yard Game With Rendezvous Spaceflight Operation Applications, Joshua A. Daughtery

Theses and Dissertations

This thesis takes the Scotland Yard board game and modifies its rules to mimic important aspects of space in order to facilitate the creation of artificial intelligence for space asset pursuit/evasion scenarios. Space has become a physical warfighting domain. To combat threats, an understanding of the tactics, techniques, and procedures must be captured and studied. Games and simulations are effective tools to capture data lacking historical context. Artificial intelligence and machine learning models can use simulations to develop proper defensive and offensive tactics, techniques, and procedures capable of protecting systems against potential threats. Monte Carlo Tree Search is a bandit-based …

Algorithm Selection Framework: A Holistic Approach To The Algorithm Selection Problem, Marc W. Chalé

Theses and Dissertations

A holistic approach to the algorithm selection problem is presented. The “algorithm selection framework" uses a combination of user input and meta-data to streamline the algorithm selection for any data analysis task. The framework removes the conjecture of the common trial and error strategy and generates a preference ranked list of recommended analysis techniques. The framework is performed on nine analysis problems. Each of the recommended analysis techniques are implemented on the corresponding data sets. Algorithm performance is assessed using the primary metric of recall and the secondary metric of run time. In six of the problems, the recall of …

Pedestrian Navigation Using Artificial Neural Networks And Classical Filtering Techniques, David J. Ellis

Theses and Dissertations

The objective of this thesis is to explore the improvements achieved through using classical filtering methods with Artificial Neural Network (ANN) for pedestrian navigation techniques. ANN have been improving dramatically in their ability to approximate various functions. These neural network solutions have been able to surpass many classical navigation techniques. However, research using ANN to solve problems appears to be solely focused on the ability of neural networks alone. The combination of ANN with classical filtering methods has the potential to bring beneficial aspects of both techniques to increase accuracy in many different applications. Pedestrian navigation is used as a …

Imitating Human Responses Via A Dual-Process Model Approach, Matthew A. Grimm

Theses and Dissertations

Human-autonomous system teaming is becoming more prevalent in the Air Force and in society. Often, the concept of a shared mental model is discussed as a means to enhance collaborative work arrangements between a human and an autonomous system. The idea being that when the models are aligned, the team is more productive due to an increase in trust, predictability, and apparent understanding. This research presents the Dual-Process Model using multivariate normal probability density functions (DPM-MN), which is a cognitive architecture algorithm based on the psychological dual-process theory. The dual-process theory proposes a bipartite decision-making process in people. It labels …

A Study On The Effects Of Mutation On Populations Using Strategies While Playing Iterative Prisoner's Dilemma, Ramses Romulus De Guzman Reyes

Theses and Dissertations

This thesis examines the effects different types of mutation and mutation rates have on populations using strategies while playing the Iterative Prisoners Dilemma (IPD). The system used in order to conduct this study was used in Leas et al. (2016), which uses genetic algorithms as a means of studying memory and its impact on populations playing IPD. For this study, experiments are organized into three different environments: Control, Static and Dynamic. The Control Environment focuses on analyzing the system and forming initial results. The Static Environment focuses on studying the effects of different rates on strategic populations playing IPD, while …

A Sandbox In Which To Learn And Develop Soar Agents, Daniel Lugo

Theses and Dissertations

It is common for military personnel to leverage simulations (and simulators) as cost-effective tools to train and become proficient at various tasks (e.g., flying an aircraft and/or performing a mission, among others). These training simulations often need to represent humans within the simulated world in a realistic manner. Realistic implies creating simulated humans that exhibit behaviors that mimic real-world decision making and actions. Typically, to create the decision-making logic, techniques developed from the domain of artificial intelligence are used. Although there are several approaches to developing intelligent agents; we focus on leveraging and open source project called Soar, to define …

Automated Conjecturing Approach For Benzenoids, David Muncy

Theses and Dissertations

Benzenoids are graphs representing the carbon structure of molecules, defined by a closed path in the hexagonal lattice. These compounds are of interest to chemists studying existing and potential carbon structures. The goal of this study is to conjecture and prove relations between graph theoretic properties among benzenoids. First, we generate conjectures on upper bounds for the domination number in benzenoids using invariant-defined functions. This work is an extension of the ideas to be presented in a forthcoming paper. Next, we generate conjectures using property-defined functions. As the title indicates, the conjectures we prove are not thought of on our …

Unified Behavior Framework For Discrete Event Simulation Systems, Alexander J. Kamrud

Theses and Dissertations

Intelligent agents provide simulations a means to add lifelike behavior in place of manned entities. Generally when developed, a single intelligent agent model is chosen, such as rule based, behavior trees, etc. This choice introduces restrictions into what behaviors agents can manifest, and can require significant testing in edge cases. This thesis presents the use of the UBF in the AFSIM environment. The UBF provides the flexibility to implement any and all intelligent agent models, allowing the developer to choose the model he/she feels best fits the experiment at hand. Furthermore, the UBF demonstrates several key software engineering principles through …

Evolutionary Artificial Neural Network Weight Tuning To Optimize Decision Making For An Abstract Game, Corey M. Miller

Theses and Dissertations

Abstract strategy games present a deterministic perfect information environment with which to test the strategic capabilities of artificial intelligence systems. With no unknowns or random elements, only the competitors’ performances impact the results. This thesis takes one such game, Lines of Action, and attempts to develop a competitive heuristic. Due to the complexity of Lines of Action, artificial neural networks are utilized to model the relative values of board states. An application, pLoGANN (Parallel Lines of Action with Genetic Algorithm and Neural Networks), is developed to train the weights of this neural network by implementing a genetic algorithm over a …

Developing An Effective And Efficient Real Time Strategy Agent For Use As A Computer Generated Force, Kurt Weissgerber

Theses and Dissertations

Computer Generated Forces (CGF) are used to represent units or individuals in military training and constructive simulation. The use of CGF significantly reduces the time and money required for effective training. For CGF to be effective, they must behave as a human would in the same environment. Real Time Strategy (RTS) games place players in control of a large force whose goal is to defeat the opponent. The military setting of RTS games makes them an excellent platform for the development and testing of CGF. While there has been significant research in RTS agent development, most of the developed agents …

Real-Time Automatic Price Prediction For Ebay Online Trading, Ilya Igorevitch Raykhel

Theses and Dissertations

While Machine Learning is one of the most popular research areas in Computer Science, there are still only a few deployed applications intended for use by the general public. We have developed an exemplary application that can be directly applied to eBay trading. Our system predicts how much an item would sell for on eBay based on that item's attributes. We ran our experiments on the eBay laptop category, with prior trades used as training data. The system implements a feature-weighted k-Nearest Neighbor algorithm, using genetic algorithms to determine feature weights. Our results demonstrate an average prediction error of 16%; …

Simulation And Visualization Of Environments With Multidimensional Time, Luther A. Tychonievich

Theses and Dissertations

This work introduces the notion of computational hypertime, or the simulation and visualization of hypothetical environments possessing multidimensional time. An overview of hypertime is provided,including an intuitive visualization paradigm and a discussion of the failure of common simulation techniques when extended to include multidimensional time. A condition for differential equations describing hypertime motion to be amenable to standard time-iterative simulation techniques is provided,but is not satisfied by any known model of physics. An alternate simulation algorithm involving iterative refinement of entire equations of motion is presented,with an example implementation to solve elastic collisions in hypertime. An artificial intelligence algorithm for …

A Unified Framework For Solving Multiagent Task Assignment Problems, Kevin Cousin

Theses and Dissertations

Multiagent task assignment problem descriptors do not fully represent the complex interactions in a multiagent domain, and algorithmic solutions vary widely depending on how the domain is represented. This issue is compounded as related research fields contain descriptors that similarly describe multiagent task assignment problems, including complex domain interactions, but generally do not provide the mechanisms needed to solve the multiagent aspect of task assignment. This research presents a unified approach to representing and solving the multiagent task assignment problem for complex problem domains. Ideas central to multiagent task allocation, project scheduling, constraint satisfaction, and coalition formation are combined to …

Parallelization Of Ant Colony Optimization Via Area Of Expertise Learning, Adrian A. De Freitas

Theses and Dissertations

Ant colony optimization algorithms have long been touted as providing an effective and efficient means of generating high quality solutions to NP-hard optimization problems. Unfortunately, while the structure of the algorithm is easy to parallelize, the nature and amount of communication required for parallel execution has meant that parallel implementations developed suffer from decreased solution quality, slower runtime performance, or both. This thesis explores a new strategy for ant colony parallelization that involves Area of Expertise (AOE) learning. The AOE concept is based on the idea that individual agents tend to gain knowledge of different areas of the search space …

Scaling Ant Colony Optimization With Hierarchical Reinforcement Learning Partitioning, Erik J. Dries

Theses and Dissertations

This research merges the hierarchical reinforcement learning (HRL) domain and the ant colony optimization (ACO) domain. The merger produces a HRL ACO algorithm capable of generating solutions for both domains. This research also provides two specific implementations of the new algorithm: the first a modification to Dietterich's MAXQ-Q HRL algorithm, the second a hierarchical ACO algorithm. These implementations generate faster results, with little to no significant change in the quality of solutions for the tested problem domains. The application of ACO to the MAXQ-Q algorithm replaces the reinforcement learning, Q-learning and SARSA, with the modified ant colony optimization method, Ant-Q. …