U.S. Department of Energy

Pacific Northwest National Laboratory


What is the problem you are trying to solve? Who are the stakeholders and why should they care about this? How do we accelerate the process of signature discovery?

Currently, most systems start from scratch each time they are used, forcing each user to work in isolation. Can we leverage the work of others to help a system learn and further increase its value for signature discovery? In addition, this system allows us to simultaneously analyze heterogeneous data, such as pictures and images, which has been a challenging and unsolved problem.


Often, a user is attempting to differentiate between items that are only distinct when a context is provided: a picture of three birds could be intended as a representation of the number "3" or an example of flying animals. By arranging items on the canvas, users implicitly communicate the latent mental model they are using, providing this necessary context. Using information theoretic measures, we use the arrangements on the screen to determine which features the user is utilizing. To aide the user, the canvas then subtly perturbs the positions on the screen to suggest arrangements that increase the mutual information between the key features and positions. The system can also suggest positions for new data that has not been considered by the user. The user's arrangement of items on the screen imbues semantic relationships among disparate items. Having canvas remember these arrangements increases the value of the data for other users. So, by effectively "crowdsourcing" new, salient, features, the canvas can help users to leverage other signatures to more rapidly establish new models.


This Canvas extends the functionality of the Signature Discovery Workbench, allowing a new level of productivity and facilitates team-based signature discovery.

Additional Staff: 
Russ Burtner, Nathan Hodas
| Pacific Northwest National Laboratory