Joint Optimization of Fidelity and Commensurability for Manifold Alignment and Vertex Assignment

• Activity Number: 672
• Type: Contributed
• Date/Time: Thursday, August 2, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Sancar Adali*+
• Keywords: matched data ; fidelity and commensurability ; hypothesis testing ; manifold alignment ; vertex correspondence ; dissimilarity representation

Abstract

Joint optimization of Fidelity and Commensurability (JOFC) was proposed as a solution to the manifold alignment problem where matched data from multiple conditions are available. It is particularly suited for problems where data is available in the form of dissimilarities. JOFC considers both fidelity of representation and preservation of commensurability when finding a common representation for multiple domain data. Fine control of tradeoff between fidelity and commensurability is an important feature of JOFC. Tuning the tradeoff parameter results in better performance for different inference tasks. We apply JOFC to a vertex correspondence problem on matched graphs and compare the performance to the Hungarian algorithm for assignment problem.