Graduate Student Talk at UMBC
Mathematics/Psychology : 401
Date & Time
May 8, 2023, 11:00 am – 11:30 am
Title: Unsupervised Machine Learning Indicators of Data and Distribution Uncertainty
Speaker: Francesca McFadden, advised by Dr. Matthias Gobbert
Abstract: Utility of Machine learning techniques is predicated on the available training data which may be accessed, measured, or simulated. Machine learning techniques utilize the point estimates from a data set to surmise the data category, predict data estimates, and select features for algorithm evaluation. There are two major types of Machine learning algorithms - Supervised versus Unsupervised. Unsupervised techniques are applied to unlabeled data. Clustering is an unsupervised machine learning technique to group similar observations in the data set into a discrete number of clusters. The talk provides an overview of clustering algorithms and then highlights an application employing the Mahalanobis distance metric. The application is a methodology using unsupervised learning to expand current approaches to recognize when there is a lack of prediction competence for a supervised machine learning model. Model competence is an indicator of how well a model is expected to perform on inputs outside of its training set. Model competency metrics enable detection of when data being processed is significantly outside the prediction space of a machine learned model. Additional current and future applications of clustering algorithms being explored are discussed.