A Framework for Interactive t-SNE Clustering

In this paper, we describe our progress in creating the framework for aninteractive application that allows humans to actively participate in a t-SNE clusteringprocess. t-SNE (t-Distributed Stochastic Neighbor Embedding) is a dimensionalityreduction technique that maps high dimensional data sets to lower dimensions that canthen be visualized for human interpretation. By prompting users to monitor outlyingpoints during the t-SNE clustering process, we hypothesize that users may be able to makeclustering faster and more accurate than purely algorithmic methods. Further researchwould test these hypotheses directly. We would also attempt to decrease the lag timebetween the various components of our application and develop an intuitive approach forhumans to aid in clustering unlabeled data. Research into human assisted clustering cancombine the strengths of both humans and computer programs to improve the results ofdata analysis.