Peri Tarr

Cognitive software--which uses machine learning and data science to enable the software to "partner" with humans and augment human abilities--is here. Cognitive software engineering poses some unique challenges to the software practitioner and research communities. One of these is embodied in the cognitive components themselves: they must be trained and tuned appropriately, so that they can understand and respond correctly to users of the cognitive application. It is, however, very difficult to train these components so that they produce good results in your application; to diagnose issues that are impeding their performance; and even to know when they are "good enough" for use. Traditionally, only machine learning and data science experts have been able to produce well-performing cognitive components successfully. Most developers do not have the required knowledge and expertise, but they are the ones creating and evolving cognitive software.

In this talk, I will describe some of the challenges that cognitive software engineering poses for the software engineering research and practitioner communities. I will also demonstrate FARCAST, a framework and tool that aims to help software engineering experts to produce well-performing cognitive components for cognitive applications. FARCAST leverages knowledge expert knowledge and insights to help reduce the time and expertise required to produce an appropriately trained model for your cognitive application.