University of Hawaii

Electrical Engineering

Atypicality for the Class of Exponential Family

Date: 2016-09-14           Add to Google Calendar
Time: 4:00pm - 5:00pm
Location: Holmes Hall 389
Speaker: Elyas Sabeti, PhD student, University of Hawaiʻi Electrical Engineering

One characteristic of the information age is the exponential growth of information, and the ready availability of this information through networks, including the internet – “Big Data.” The question is what to do with this enormous amount of  information. One possibility is to characterize it through statistics – think averages. Our perspective is the opposite, namely that most of the value in the information is in the parts that deviate from the average, that are unusual, atypical. The rest is just background noise. Take art: the truly valuable paintings are those that are rare and atypical. The same could be true for scientific research and entrepreneurship. Take online collections of photos, such as Most of the photos are rather pedestrian snapshots and not of interest to a wider audience. The photos that of interest are those that are unique. Flickr has a collection of photos rated for ’interestingness,’ and one can notice that those photos are indeed very different from typical photos. They are atypical.

Atypicality is a new concept that uses a code length based deviation from the norm to find the interesting rare events. In a previous work we have developed an information theoretic approach for binary data. Then in the two other works, we came up with an extension to the real-valued models for Gaussian and vector Gaussian cases. This talk is about a generalization of our real-valued model to the class of exponential family, in which many properties of atypicality will be discussed.