Seminar Series | Graduate Student Colloquiums
144 Baker Systems
1971 Neil Avenue
Columbus, OH 43210
Presenter: Ron Gantt
Advisor: Dr. Mike Rayo
Committee: Dr. Emily Patterson, and Dr. David Woods
Title: Coordination Across Boundaries: Coordination Costs, New Forms of Brittleness, and the Multi-Party Dilemma in the Construction Industry
Abstract: Organizations in numerous domains are increasingly relying on external parties in order to achieve their goals, creating increasingly complex tangled networks of parties that must coordinate in order to be successful. However, a growing body of research has identified that coordination across boundaries may introduce new forms of challenge and potential failure. This research seeks to use the construction industry, a domain that historically relies on coordination across boundaries to identify what system architectures support or hobble joint activity. Using an extension of the critical decision method of cognitive interviewing with domain experts in the data center construction industry, this research hopes to identify and analyze a corpus of cases where coordination across boundaries works, where it is strained, and where it breaks down. The patterns identified will be compared with patterns found in other domains, such as critical digital services and transportation.
Presenter: Maha Yazbeck
Advisor: Dr. Ted Allen
Committee: Dr. Samantha Krening, Dr. Bill Notz, Dr. Marat Khafizov
Title: Innovative Experimentation and analysis methods for image analysis and generation in pipeline and neuroscience.
Abstract: In this research, we address two critical applied challenges: pipeline corrosion assessment and brain connectivity analysis in ADHD populations. For the first problem, pipelines in the United States span approximately 2.3 million miles, and corrosion-related incidents incurred significant economic losses. Leveraging artificial intelligence (AI), particularly semi-physics-based simulations and generative AI (Variational Autoencoder or VAE), we aim to predict corrosion progression, generate images and infer boundary conditions for pipeline inspection. This involves optimizing simulation parameters using forward inverse space filling design and accelerating parameter search through neural network transformations. Additionally, VAE is employed to generate synthetic corrosion images based on corrosion attributes, expanding the dataset for classification algorithms. For brain connectivity analysis in ADHD, we propose a novel validation method for order statistics based on Monte Carlo simulations to assess EEG signal deviations. The research combines advanced AI techniques with simulation and statistical approaches to address these complex real-world challenges effectively.
Presenter: Ryan Gifford
Advisor: Samantha Krening
Committee: Mike Rayo, Sachin Jhawar, Eric Fosler-Lussier
Title: CNN Trees for Explainable Time Series Classification
Abstract: Recent advances in the field of time-series classification involve the use of Deep Learning due to high performance. However, Deep Learning based models often come at the cost of being black box as we have little to no insight into their underlying decision processes. In high stakes environments model interpretability is critical to successful performance and the adoption of black box models should be cautioned against. In this research we propose CNN Trees for time series classification. CNN Trees are inherently explainable and combine representation learning with the symbolic structure of the decision tree.