Conjoint Audiogram Estimation via Gaussian Process Classification

In traditional audiometry, a clinician seeks to estimate her patient’s auditory response through the sequential delivery of various individual tests. These tests are treated as independent and correlation is assessed after each individual test has been completed, resulting in a diagnosis. Treating tests as independent impedes both accuracy and efficiency by ignoring correlations in conditions known to influence physiological response, for instance age, genetics, and exposure to noise. This thesis advances the existing framework for audiometry via Gaussian Processes by allowing for the estimation of audiogram thresholds for both ears simultaneously. The resulting model estimates both correlated and uncorrelated right- and left-ear audiograms with higher efficiency than was previously achievable. This work lays a foundation for building further estimation between discrete psychometric spaces