Interactive Machine Learning|Letting Users Build Classiers

According to standard procedure, building a classi er is a fully automated process that follows data preparation by a domain ex-pert. In contrast, interactive machine learn-ing engages users in actually generating the classi er themselves. This oers a natu-ral way of integrating background knowledge into the modeling stage|so long as interac-tive tools can be designed that support e-cient and eective communication. This pa-per shows that appropriate techniques can empower users to create models that compete with classiers built by state-of-the-art learn-ing algorithms. It demonstrates that users| even users who are not domain experts| can often construct good classiers, with-out any help from a learning algorithm, us-ing a simple two-dimensional visual interface. Experiments demonstrate that, not surpris-ingly, success hinges on the domain: if a few attributes can support good predictions, users generate accurate classiers, whereas domains with many high-order attribute in-teractions favor standard machine learning techniques. The future challenge is to achieve a symbiosis between human user and machine learning algorithm. 1