A Nonmetric Embedding Approach to Testing for Matched Pairs

• Activity Number: 354
• Type: Contributed
• Date/Time: Tuesday, July 31, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Computing
• Author(s): Brent Castle*+ and Michael Trosset and Carey E Priebe
• Companies: Indiana University and Indiana University and The Johns Hopkins University
• Keywords: three-way multidimensional scaling ; multi-view learning ; data fusion ; multiple modalities ; heterogeneous data

Abstract

We consider the problem of matched pair hypothesis testing, i.e., the problem of determining whether or not a pair of disparate feature vectors correspond to a common object. Our approach relies on measures of pairwise dissimilarity in the respective feature spaces. We use three-way nonmetric multidimensional scaling to embed a training set of matched pairs in a low-dimensional Euclidean space, then embed test pairs by a corresponding out-of-sample technique. The Euclidean distance between the points obtained by embedding a test pair is used as a statistic to test whether or not the pair is matched. We demonstrate our methodology matching photos of faces taken under two conditions.