Seminar: ISE graduate student colloquiums

All dates for this event occur in the past.

144 Baker Systems
144 Baker Systems
1971 Neil Avenue
Columbus, OH 43210
United States


Title: Designing simulation-based educational activities using augmented reality and off-line optimal experimental design for gaming

Presenter: Olivia Hernandez

Committee: Theodore T. Allen (advisor), Emily Patterson

Abstract: The primary objective is to help decision-makers to select the most advantageous low consequence learning scenarios and to analyze the associated experiment results. This research considers both adversarial gaming and education contexts (which maybe adversarial to an extent). In both contexts, participants must learn the rewards from different action options through low consequence experimentation. For example, medical students can perform symptom recognition activities using augmented reality (AR). By designing these low consequence experiments and analysis methods carefully, effective learning and participation in on-the-job or real game tasks is enabled. Specifically, for training medical students, we investigate how realistic cues (not drawings) from AR can assist medical students to make diagnoses which rely on dynamic physical cues.  


Title: Large Scale Optimal Classification Trees and Cyber Security Prioritization and Control

Presenter: Enhao Liu

Committee: Theodore T. Allen (advisor), Cathy Xia and Guzin Bayraksan

Abstract: Cyber vulnerability management is difficult because there are tens of thousands of distinct vulnerabilities distributed across large numbers of hosts including PCs, laptops, printers, or even die casting or exercise machines. Dealing with these vulnerabilities requires prioritizing patching or remediation activities, task which can clearly benefit by evaluating which specific vulnerabilities are likely to be exploited in a specific organization. In this research, we want to build systems based on machine learning methods that enable cyber security IT stuffs to better analyze, detect, prioritize and quickly remedy the most important vulnerabilities, thereby saving millions of dollars for organizations. It is also important to understand machine learning models. The ensemble learning methods or neural networks are “black box” and they are less interpretable than linear models or a single decision tree. We intend to develop a convergent sampling framework to enable optimal classification trees for big data with proven finite sample and asymptotic results, thereby efficiently addressing interpretability concerns.


Title: Hyperparameter Optimization for Deep Learning Neural Networks Using Simulation Optimization

Presenter: Mehdi Mashayekhi

Committee: Theodore T. Allen (advisor), Cathy Xia and Guzin Bayraksan

Abstract: Deep Neural Networks (DNNs) have a miscellaneous range of applications. From recovering  damaged WAV audio files to forecasting day-ahead electricity loads. DNNs are accessible, easy to use and accurate. However, there is uncertainty in the output of DNNs. This uncertainty is mainly due to the dataset and the model structure. To be more specific, there are some parameters which need to be set by the operator before running the model. These parameters are called Hyperparameters. In this research, we try to introduce a globally convergent Simulation Optimization algorithm to reduce the uncertainty of the output while getting good outputs.


Title: Discovering short and long-term effects of chronic pressure on adaptive systems

Presenter: Kati Walker

Committee: Michael Rayo (advisor), David Woods, Emily Patterson

Abstract: Organizational pressures are both omnipresent and largely invisible to the organization and the practitioners within it. Whereas these pressures are necessary (Dekker, 2006), they also contribute to a number of system design decisions and behaviors that, together, threaten system safety. One focus of the resilience engineering field has been to develop ways to predict the effects of change on the resilience of the system, and particularly to use those findings to recognize and anticipate paths to failure (Woods and Hollnagel, 2006). The existing research has revealed high-level patterns in how systems succeed or move toward failure (e.g., Woods & Branlat, 2011). There is a need to extend these patterns by finding specific and observable behaviors that manifest as a result of strong production and financial patterns. This research uses the Systemic Contributors and Adaptations Diagramming method to analyze a corpus of accidents across diverse industries, such as medicine, aerospace, and water management. The goal is to identify system pressures by discovering their short and long-term sequelae that appear across industries.


 

Category: Seminars