Browsing by Subject "machine learning"
Now showing 1 - 8 of 8
Results Per Page
Sort Options
Item A Digital Image Library: Making it possible with Facial Recognition(Texas Digital Library, 2023-05-17) Prud'homme, Max; Peta, LokeshEnsuring the discovery and preservation of digital archival assets is an important aspect of digital curation work at the Oklahoma State University Library. In the fall of 2023, the university archives resumed their machine learning work after conducting a successful pilot project that explored the use of facial recognition techniques to curate a high-value archival collection. With support from Library Administration, the digital archives are moving forward with the development of a dynamic search engine, using machine learning, to improve the predictability and performance for searching thousands of digital assets. To achieve this, the team is constructing a model that is easily trainable and an interactive application to search images more efficiently. With consideration to scalability and sustainability, the facial recognition technology used in the pilot project is being extended to a larger and more diverse dataset of face images. The presenters propose to showcase the project flow, context, planning, design and architecture in a demonstration/tutorial-like presentation. They will address challenges and initial feedback, with a particular focus on scalability, sustainability, as well as ethical issues associated with facial recognition technology.Item A Recommendation System for Preconditioned Iterative Solvers(2011-02-22) George, ThomasSolving linear systems of equations is an integral part of most scientific simulations. In recent years, there has been a considerable interest in large scale scientific simulation of complex physical processes. Iterative solvers are usually preferred for solving linear systems of such magnitude due to their lower computational requirements. Currently, computational scientists have access to a multitude of iterative solver options available as "plug-and- play" components in various problem solving environments. Choosing the right solver configuration from the available choices is critical for ensuring convergence and achieving good performance, especially for large complex matrices. However, identifying the "best" preconditioned iterative solver and parameters is challenging even for an expert due to issues such as the lack of a unified theoretical model, complexity of the solver configuration space, and multiple selection criteria. Therefore, it is desirable to have principled practitioner-centric strategies for identifying solver configuration(s) for solving large linear systems. The current dissertation presents a general practitioner-centric framework for (a) problem independent retrospective analysis, and (b) problem-specific predictive modeling of performance data. Our retrospective performance analysis methodology introduces new metrics such as area under performance-profile curve and conditional variance-based finetuning score that facilitate a robust comparative performance evaluation as well as parameter sensitivity analysis. We present results using this analysis approach on a number of popular preconditioned iterative solvers available in packages such as PETSc, Trilinos, Hypre, ILUPACK, and WSMP. The predictive modeling of performance data is an integral part of our multi-stage approach for solver recommendation. The key novelty of our approach lies in our modular learning based formulation that comprises of three sub problems: (a) solvability modeling, (b) performance modeling, and (c) performance optimization, which provides the flexibility to effectively target challenges such as software failure and multiobjective optimization. Our choice of a "solver trial" instance space represented in terms of the characteristics of the corresponding "linear system", "solver configuration" and their interactions, leads to a scalable and elegant formulation. Empirical evaluation of our approach on performance datasets associated with fairly large groups of solver configurations demonstrates that one can obtain high quality recommendations that are close to the ideal choices.Item Developing intelligent agents for training systems that learn their strategies from expert players(Texas A&M University, 2005-11-01) Whetzel, Jonathan HuntComputer-based training systems have become a mainstay in military and private institutions for training people how to perform certain complex tasks. As these tasks expand in difficulty, intelligent agents will appear as virtual teammates or tutors assisting a trainee in performing and learning the task. For developing these agents, we must obtain the strategies from expert players and emulate their behavior within the agent. Past researchers have shown the challenges in acquiring this information from expert human players and translating it into the agent. A solution for this problem involves using computer systems that assist in the human expert knowledge elicitation process. In this thesis, we present an approach for developing an agent for the game Revised Space Fortress, a game representative of the complex tasks found in training systems. Using machine learning techniques, the agent learns the strategy for the game by observing how a human expert plays. We highlight the challenges encountered while designing and training the agent in this real-time game environment, and our solutions toward handling these problems. Afterward, we discuss our experiment that examines whether trainees experience a difference in performance when training with a human or virtual partner, and how expert agents that express distinctive behaviors affect the learning of a human trainee. We show from our results that a partner agent that learns its strategy from an expert player serves the same benefit as a training partner compared to a programmed expert-level agent and a human partner of equal intelligence to the trainee.Item Measurement to Intelligence: Feature Extraction, Modeling and Predictive Analysis of Asymmetric Conflict Events(2014-06-06) George, Stephen MThe conflict events that comprise asymmetric warfare are a primary killer of both combatants and civilians on the modern battlefield. Improvised explosive devices (IED) and direct fire (DF), the most common of these attacks, claim thousands of lives as conventional and unconventional forces clash. Computer-based predictive analysis can be used to identify locations that are useful for these events, potentially providing the awareness needed to disrupt or avoid attacks before they are launched. In this dissertation, I propose an analytical framework for predictive analysis of asymmetric conflict events. This framework incorporates a tactics-aware system model based on attacker roles that is populated with a set of geomorphometric and visibility-constrained features describing terrain and proximity to necessary supporting structures. Features that identify and assess the utility of terrain for use by risk-averse attackers are important contributors to the model. Statistical learning is used to extract spatially and temporally constrained tactical patterns. These patterns are then used to predict the utility of future or unvisited locations for conflict events. Major contributions of this dissertation include: (1) A concise, accurate feature representation of conflict events in non-urban environments; (2) A system model based on attacker roles that captures the tactical patterns of conflict events; (3) Accurate conflict event classification algorithms that support predictive analysis; and (4) A novel method for detecting and describing features that support risk-averse attackers. The framework has been implemented and tested on real-world IED and DF data collected from the conflict in Afghanistan in 2011-2012. Several learning techniques are assessed using two dimensionality reduction schemes under a variety of spatial, temporal and combined constraints. A resource-unconstrained version of the framework accurately predicts conflict events across a wide range of terrain types and over the 19 months covered by available data. A limited version of the framework that assumes less computational capability provides useful predictive analysis that can be performed in mobile and resource constrained environments.Item Overcoming Clonal Interference in Escherichia coli Using Genderless High Frequency Recombination Strains(2014-04-16) Winkler, JamesAdaptive laboratory evolution (ALE) is a powerful tool for strain improvement, and has been applied successfully to improve a range of desirable phenotypes in model organisms through continuous cultivation under a selective pressure of interest. Despite its demonstrable utility, one limiting factor for the effectiveness of ALE is competition between beneficial mutants that exist contemporaneously within an evolving population. This phenomenon of clonal interference arises from the fact that the majority of microbes are obligate asexual organisms that cannot exchange DNA between cells. Mutants that arise must therefore compete for resources until the fittest mutant drives the others to extinction. The resulting loss of genetic information from the population slows the overall rate of adaptation, and decreases the amount of information that can be extracted from a given ALE experiment. To overcome these limitations, we have developed a novel in situ mating system based on the F plasmid to allow continuous DNA exchange between E. coli cells in liquid culture, allowing mutants to potentially combine their mutations into a single genetic background. The utility and limitations of an existing recombination method, genome shuffling, are also explored to demonstrate the advantages of this new method. The design and initial testing of the in situ mating system is first validated, and the system is used for a subsequent evolution experiment under osmotic stress to validate the industrial applicability of the mating system. Adaptive mutants generated in the course of these experiments are then used to test whether tolerant mutants can be formed via conjugation. Finally, additional side projects focusing on strain or population characterization tools are discussed, followed by recommendations for future work.Item Playing Hide-and-Seek with Spammers: Detecting Evasive Adversaries in the Online Social Network Domain(2012-10-19) Harkreader, Robert ChandlerOnline Social Networks (OSNs) have seen an enormous boost in popularity in recent years. Along with this popularity has come tribulations such as privacy concerns, spam, phishing and malware. Many recent works have focused on automatically detecting these unwanted behaviors in OSNs so that they may be removed. These works have developed state-of-the-art detection schemes that use machine learning techniques to automatically classify OSN accounts as spam or non-spam. In this work, these detection schemes are recreated and tested on new data. Through this analysis, it is clear that spammers are beginning to evade even these detectors. The evasion tactics used by spammers are identified and analyzed. Then a new detection scheme is built upon the previous ones that is robust against these evasion tactics. Next, the difficulty of evasion of the existing detectors and the new detector are formalized and compared. This work builds a foundation for future researchers to build on so that those who would like to protect innocent internet users from spam and malicious content can overcome the advances of those that would prey on these users for a meager dollar.Item Reinforcement Learning for Active Length Control and Hysteresis Characterization of Shape Memory Alloys(2010-01-16) Kirkpatrick, Kenton C.Shape Memory Alloy actuators can be used for morphing, or shape change, by controlling their temperature, which is effectively done by applying a voltage difference across their length. Control of these actuators requires determination of the relationship between voltage and strain so that an input-output map can be developed. In this research, a computer simulation uses a hyperbolic tangent curve to simulate the hysteresis behavior of a virtual Shape Memory Alloy wire in temperature-strain space, and uses a Reinforcement Learning algorithm called Sarsa to learn a near-optimal control policy and map the hysteretic region. The algorithm developed in simulation is then applied to an experimental apparatus where a Shape Memory Alloy wire is characterized in temperature-strain space. This algorithm is then modified so that the learning is done in voltage-strain space. This allows for the learning of a control policy that can provide a direct input-output mapping of voltage to position for a real wire. This research was successful in achieving its objectives. In the simulation phase, the Reinforcement Learning algorithm proved to be capable of controlling a virtual Shape Memory Alloy wire by determining an accurate input-output map of temperature to strain. The virtual model used was also shown to be accurate for characterizing Shape Memory Alloy hysteresis by validating it through comparison to the commonly used modified Preisach model. The validated algorithm was successfully applied to an experimental apparatus, in which both major and minor hysteresis loops were learned in temperature-strain space. Finally, the modified algorithm was able to learn the control policy in voltage-strain space with the capability of achieving all learned goal states within a tolerance of +-0.5% strain, or +-0.65mm. This policy provides the capability of achieving any learned goal when starting from any initial strain state. This research has validated that Reinforcement Learning is capable of determining a control policy for Shape Memory Alloy crystal phase transformations, and will open the door for research into the development of length controllable Shape Memory Alloy actuators.Item Scaling Up Reinforcement Learning without Sacrificing Optimality by Constraining Exploration(2012-12-05) Mann, Timothy 1984-The purpose of this dissertation is to understand how algorithms can efficiently learn to solve new tasks based on previous experience, instead of being explicitly programmed with a solution for each task that we want it to solve. Here a task is a series of decisions, such as a robot vacuum deciding which room to clean next or an intelligent car deciding to stop at a traffic light. In such a case, state-of-the-art learning algorithms are difficult to employ in practice because they often make thou- sands of mistakes before reliably solving a task. However, humans learn solutions to novel tasks, often making fewer mistakes, which suggests that efficient learning algorithms may exist. One advantage that humans have over state- of-the-art learning algorithms is that, while learning a new task, humans can apply knowledge gained from previously solved tasks. The central hypothesis investigated by this dissertation is that learning algorithms can solve new tasks more efficiently when they take into consideration knowledge learned from solving previous tasks. Al- though this hypothesis may appear to be obviously true, what knowledge to use and how to apply that knowledge to new tasks is a challenging, open research problem. I investigate this hypothesis in three ways. First, I developed a new learning algorithm that is able to use prior knowledge to constrain the exploration space. Second, I extended a powerful theoretical framework in machine learning, called Probably Approximately Correct, so that I can formally compare the efficiency of algorithms that solve only a single task to algorithms that consider knowledge from previously solved tasks. With this framework, I found sufficient conditions for using knowledge from previous tasks to improve efficiency of learning to solve new tasks and also identified conditions where transferring knowledge may impede learning. I present situations where transfer learning can be used to intelligently constrain the exploration space so that optimality loss can be minimized. Finally, I tested the efficiency of my algorithms in various experimental domains. These theoretical and empirical results provide support for my central hypothesis. The theory and experiments of this dissertation provide a deeper understanding of what makes a learning algorithm efficient so that it can be widely used in practice. Finally, these results also contribute the general goal of creating autonomous machines that can be reliably employed to solve complex tasks.