Fusion Learning by Individual-to-Clique (FLIC): Efficient Approach to Enhancing Individual Inference Through Adaptive Combination of Confidence Distributions

• Activity Number: 400
• Type: Invited
• Date/Time: Tuesday, August 11, 2015 : 2:00 PM to 3:50 PM
• Sponsor: General Methodology
• Author(s): Minge Xie* and Regina Y. Liu and Jieli Shen
• Companies: Rutgers University and Rutgers University and Rutgers University
• Keywords: Big data ; Combining information ; Confidence distribution ; Fusion Learning ; Joint inference ; Split-and-conquer

Abstract

Learning from multiple studies can often be fused together to yield a more effective overall inference than individual studies alone. Such effective fusion learning is of vital importance, especially in light of the trove of data nowadays collected routinely from various sources in all domains and at all time. We present a new approach, named "Fusion Learning by Individual-to-Clique (FLIC)," to enhancing inference of an individual study through adaptive combination of confidence distributions obtained from its clique (namely peers of similar studies). Roughly speaking, FLIC begins with obtaining inference for each individual study, then adaptively forming a clique, and finally obtaining a combined inference from the clique. FLIC can be performed without accessing the entire data; thus allow the so-called split & conquer approach to be implemented on individual studies and reduce substantially computational expense. Drawing inference from the clique allows borrowing strength from similar studies to enhance the inference efficiency for individual studies. We also provide supporting theories for FLIC and its applications in personalized medicine and financial profiling of companies.