Visual Machine Learning for Anomaly and Rare Category Detection

Machine learning is widely used for classification tasks.  Generally speaking, this approach aims to identify the most representative and self-similar classes within a dataset.  In many cases, however, these well-represented classes are what users of a system find interesting.  For example, security analysts looking for computer-controlled social network accounts are looking for relatively rare behaviors which are anomalous in some way.  Similarly, physicians looking for arrhythmias within large ECG data streams are looking for rare but diagnostically significant heartbeat patterns.

In these projects, we’ve developed visualization-based interfaces which couple closely with machine learning algorithms to aid in this type of task.  This includes work with deep learning models for heart arrhythmia detection which won an Honorable Mention Award at ACM CHI 2018.

Publications:

Ke Xu, Shunan Guo, Nan Cao, David Gotz, Aiwen Xu, Huamin Qu, Zhenjie Yao, Yixin Chen. ECGLens: Interactive Visual Exploration of Large Scale ECG Data for Arrhythmia Detection. ACM CHI (2018). Honorable Mention Award
[A link to this article will be provided when available online.  In the meantime, a preprint is available here.]

Hanfei Lin, Siyuan Gao, David Gotz, Fan Du, Jingrui He, and Nan Cao. RCLens: Interactive Rare Category Exploration and Identification. IEEE Transactions on Visualization and Computer Graphics (Published online on June 6, 2017. To appear in print.).
[This article is available through the journal’s website. View the article by clicking here.]