|
Abstract:
|
We build an ensemble machine learning method to integrate and optimize existing facial recognition algorithms. SuperLearner is a general loss-based ensemble learning method designed to find the optimal combination of an a priori specified collection of prediction algorithms by minimizing the cross-validated risk with ideal asymptotic oracle performance. Using the AT&T Laboratories Cambridge database of faces we deploy a nested SuperLearner for optimal prediction across both featurization and classification. Our SuperLearner estimator considers featurization and classification methods including principal component and linear discriminant analyses, logistic regression, polynomial splines, support vector machines and deep learning. Individual featurization and classification algorithms and our SuperLearner estimator are compared using various classification diagnostics.
|
