Auraria campus
Jacob Johns’s PhD in Applied Mathematics Research Proposal
Tuesday, December 8 · 9:30 am–11:30 am
at Student Commons 4017

Title: Anomaly Detection Methods and Causal Learning with Spatiotemporal Considerations Abstract: This proposal will focus on three sequential projects that use anomalies as a treatment in casual inference models. The first project will provide the mathematical and theoretical background needed to adapt anomaly detection methods to a causal framework. This allows the effect of anomalies on a predictive model to be measured on an individual, conditional, and population level. Next, the stable unit treatment value assumption (SUTVA) is relaxed to leverage additional network interference information from spatiotemporal sources. We propose a novel algorithm, utilizing Kolmogorov Arnold Networks (KANs) to learn both the network interference pattern and identify anomalies in the same framework. Finally, non-binary treatments and the SUTVA are simultaneously addressed, again using KANs. Specially, the learned network interference is used in a graph convolutional neural network with non-binary output nodes.
Source: cu-denver · updated automatically. Times are Denver time. Doors, prices and lineups change — the ticket link is the truth.