At the University of Florida Data Studio we are investigating research questions across the full range of the data pipeline, from data sourcing and acquisition to visualization and interaction. One direction that we are excited about is called Speed Labeling. We are developing a suite of tools with recommender systems to (1) accelerate the creation of gold-standard data in a fast and fair way, and (2) and enable the interactive investigation of large models. We are interested in using novel interfaces such as large tactile devices, mobile watches, and BCI interfaces paired with interaction systems designed and adapted to the user. Given mixed modality interfaces, we plan to detect (a) label fatigue, (b) bias), and (3) systematic errors. We are developing methods to recommend adapting new interfaces to increase performance. We additionally look to study multi-user and super super-user capability. We will use this collection of tools to further explore and edit facts in large language models.
We are developing a suite of tools with recommender systems to accelerate the creation of gold standard data. Interfaces include Tactile devices, BCI interfaces, etc. We also develop algorithms to recomment the interfaces to use, the labels to label.
We have a few exciting research questions. Can we save time and money for generating gold standard data? Can we ask users to label in transformed spaces? Visualize text as visual, audio, or haptic input devices. Can we ask users to play games to accomplish various tasks?
Stay tuned for more updates on this project.