Saleema Amershi

While developers skilled in statistical machine learning have been successful in building intelligent systems to enhance human productivity and capabilities with large unstructured data sets, a fundamental limitation of relying on developers to provide these capabilities is that developers cannot possibly foresee the countless variety of distinctions end-users might want to make within large datasets in pursuit of their every day goals. A promising solution, therefore, is to enable people to interactively train machine learning systems themselves. Recent work has shown that we can create end-user interactive machine learning systems for specific applications. However, a better understanding is needed of how to design effective interaction with interactive machine learning systems. This project examines answers to this question, aiming to broaden interaction with large unstructured data and to accelerate the integration of intelligent computing into our everyday lives. [UIST 2009 (pdf)]

Education in developing regions often suffers from the lack of critical resources. A recent solution to the limited availability of compupters is multiple mouse single display groupware systems for entire classrooms. However, most of these systems have been limited to point-and-click based activities. In this research, we explore multiple mouse-based text entry techniques to enable richer educational activities.[CSCW 2010 - Best Paper Nominee (pdf)] [mov]

This project combines self-supervised information extraction (IE) techniques with a mixed initiative interface designed to encourage communal content creation (CCC). Since IE and CCC are each powerful ways to produce large amounts of structured information they have been studied extensively - but only in isolation. By combining the two methods in a virtuous feedback cycle, we aim for substantial synergy. [CHI 2009 - Best Paper Nominee (pdf)] [AAAI 2008 Senior Papers Track (pdf)]

The APEX system takes a human-in-the-loop machine learning approach to help users extract high-level events from low-level RFID data stored in relational databases. APEX automatically searches databases for high-level events in the form of statistical patterns which it presents to the user for iterative refinement and then stores for future sensor-based application use. APEX is intended to make interaction with databases more accessible to end users such as sensor-based application developers.

Web search is often viewed as a solitary task; however, there are many situations in which groups of people gather around a single computer to jointly search for information online. CoSearch is a system we developed that leverages devices cheap and ubiquitous in the environment, such as multiple mice and mobile phones, in order to facilitate co-located collaborative Web search around a shared PC.
[CHI 2008 (pdf)] [CHI 2008 Workshop on HCI for Community and International Development (pdf)] [CHI 2008 Workshop on Sensemaking (pdf)] [mov]

Two of the most cited difficulties of developing user models for intelligent interfaces are the laborious effort required by application designers to construct models, and the limited transferability of those models across applications. In this research we designed and evaluated a machine learning based framework for building user models that reduces the development costs traditionally associated with user modeling.[Journal of Educational Data Mining 2009 (pdf)] [IUI 2007 (pdf)] [AAAI 2007 Nectar Track (pdf)] [ITS 2006 (pdf)]

Educational games can induce a wide range of emotions, and so recognizing specific emotions may be valuable for an intelligent system that aims to adapt to varying student needs so as to improve learning. In this research, we investigated ths use of unsupervised machine learning for identifying biometric expressions of affective reactions exhibited by students interacting with an educational game. [ITS 2006 Workshop on Motivational and Affective Issues in ITS (pdf)]

Interactive algorithm visualizations (AVs) are powerful tools for teaching and learning concepts that are difficult to describe with static media alone. However, while countless AVs exist, their widespread adoption by the academic community has not occured due to the usability problems and mixed results of pedagogical effectiveness reported in the (AV) and education literature. In this research, we present a taxonomy of goals for designing interactive AVs that address these problems. We also describe our own experiences designing and evaluation a set of interactive AVs for learning artificial intelligence. [Interacting with Computers - The Interdisciplinary Journal of HCI 2008 (pdf)] [ITiCSE 2005 (pdf)]

AIspace is an ongoing, collaborative research project centered around a set of interactive algorithm visualization tools for learning about artificial intelligence algorithm. [http://www.aispace.org]