Often there is a mismatch between the expectations a developer has for an intelligent system and the expectation the end-user has. My research is focused on designing and evaluating interactive machine learning systems particularly in the context of model personalization approaches (like active learning) which can learn to model a concept as the end-user percieves it instead of relying solely on how the developer percieves it. This has important implications for how we approach research in activity tracking and affective computing where it is important for the learned concept (e.g.: activity, mood) to be aligned with the user's perception of it.
Often when we evaluate new algorithms to machine learning or artificial intelligence, we focus on whether or not it the algorithm did better than its predecessor on a benchmark test set (or whether it could be said that the algorithm learned the concept at all), but what about validation on a real task in the natural environment? Users live varied lifestyles with different goals and levels of technological literacy. My research aims to uncover how people experience, understand, and use intelligent systems and how that knowledge can be used to better inform the development by bringing the user experience to the focus in the design and validation of the underlying algorithms as well as the user interface.. This UX research can help identify new directions for the design and validation of machine learning and AI algorithms.
Often when we develop new technologies it is difficult to imagine all the ways they can be used and be of use. Intelligent systems are no exception. My work aims to understand how we can design intelligent systems to be flexible and adaptable so that they can be appropriated for new domains and applications. In particular, I am interested in exploring the design of interactive machine learning systems which allow the users to create entirely custom models from scratch using their domain expertise.