The weekly SILO Seminar Series is made possible through the generous support of the 3M Company and its Advanced Technology Group


with additional support from the Analytics Group of the Northwestern Mutual Life Insurance Company

Northwestern Mutual

Network-based whole-brain representational similarity learning

Urvashi Oswal & Blake Mason, Graduate Students - UW-Madison

Date and Time: Feb 08, 2017 (12:30 PM)
Location: Orchard room (3280) at the Wisconsin Institute for Discovery Building



Speaker 1: Urvashi Oswal
Title: Network-based whole-brain representational similarity learning

Abstract: Technologies such as functional magnetic resonance imaging (fMRI) provide huge amounts of data that could help improve our understanding of the human brain but they are often plagued by many complications, including noise, high-dimensionality, strong and unknown statistical correlations. In this talk I will present a new tool, called Network representational similarity analysis (NRSA), developed to tackle some of these issues with the aim of understanding how different regions of the brain function in concert to process complex information. NRSA is a whole-brain approach to discovering arbitrarily structured brain networks (possibly widely distributed and non-local) that encode similarity information. This tool has enabled discovering such structure for cases where the interesting cortical regions and networks have proven elusive.

Speaker 2: Blake Mason
Title: Metric Learning from Comparative Judgments

Abstract: We present ongoing work on learning sparse and low rank metrics from comparative judgments. This problem takes inspiration from psychology studies where researchers commonly wish to learn which items people find similar and why. As it is difficult for humans to provide fine grained information, a standard query is of the form “is item i more like j or k?”.

In this project, we extend previous work analyzing ordinal embedding from comparative judgments to the case where auxiliary information, in the form of feature vectors, are known for each item. Under a generative data model, we demonstrate that it is possible to learn the correct metric from noisy similarity judgments. We will give prediction error bounds for two different cases and describe a simple method to learn the correct metric from the judgments. This is joint work with Lalit Jain and Robert Nowak.