User Centric Hypothesis Definition
Active learning methods of today have failed to live up to their promise of revolutionizing machine learning by combining human and machine into a more powerful machine learning system. Instead, approaches have reduced the human component into a rote labeling machine with limited ability to intervene or drive the learning process, causing the user to disengage. Instead of this, a more developed and sophisticated interaction for active learning that gives the user back some control during the training process, allows the user to interact intuitively with the model, and better helps the user understand how the model is changing as a result of their interactions is needed. Broadly, AIM User Centric Hypothesis Definition (UCHD) is concerned with how to help users understand complex dynamic or streaming data.
Approach
Our approach for a solution to this difficult problem is to build the most succinct visualization possible, by leaving out information that is not relevant to the user’s current task at hand. Furthermore, the desire is to build a system supporting a human-machine partnership that is more effective than human or machine alone. This is done by carefully designing a visualization that allows a user with expert knowledge to influence the underlying machine learning model in an elegant and straightforward manner, without requiring the user to have expert knowledge of the underlying analytical model.
Benefit
This research will benefit multiple communities and stakeholders. It will introduce a new approach for active learning that will help the paradigm live up to the full potential of human-machine teaming. This will be very beneficial to the machine learning community. Novel algorithms and techniques for storyline visualization that scale in both time and space dimensions, enabling storyline visualization techniques to be used in many new application domains will also be developed. Communicating model provenance is a crucial missing component for semantic interaction, and doing so with storylines is intuitive and effective.
[Poster] Design Alternatives for Storyline Visualization of Multivariate Time Series
Patterns, relationships, and anomalies in time-varying data are revealed through exploratory visual analytics using storylines. Each entity is represented as a line, and time is encoded on the horizontal axis. When entities are similar at a given time point, their storylines come together, otherwise they are drawn apart. This poster explores some alternative algorithms for determining the y-coordinate of the storylines.
