Summary

In this position I studied and designed the elicitation of task information and task context, so it could be fed to an NLP model with the goal of autonomously visualizing a user's mental model.

Description

I had the opportunity to intern under the leadership of Dr. Jayde King in the 711th Human Performance Wing on "Agents for Co-Training and Knowledge Capture." This is a selective and prestigious opportunity provided by the Oak Ridge Institute for Science and Education (ORISE). "The 711th Human Performance Wing (711 HPW) is a unique combination of the Airman Systems Directorate (RH) and the US Air Force School of Aerospace Medicine (USAFSAM). The synergies of combining ideas, resources and technologies of these units position the 711 HPW as a world leader in the study and advancement of human performance." These efforts support the Department of Defense by advancing knowledge in Human-Machine Co-Learning and Human Factors of Warfighters. Pathfinder (source of image) was one technology we used to take mental model unstructured text data, and visualize it as a network (see image). Other technologies included BERTnet and GraphGPT, but were found to be insufficient (a presentation about these technologies is available on request). Unstructured text was converted to similarity ratings, using a covariance method in Python. Once similarity ratings were formatted into a similarity matrix, it could be passed to the Pathfinder software. The deliverables were presented during the AFRL poster session and Repperger presentation session. Human Factors Skills: Human-AI Co-Learning, Task Analysis, Mental Models, MM Representation (JPathfinder Networks), Autonomous Elicitation Methods, Interpreting Unstructured Text, Validating Autonomous Knowledge RepresentationData Science Skills: NLP, Proximity Measures (Minkowski, Cosine), Data Mining, Triples, Knowledge Graphs, GraphGPT, BERT(net), Tokenization & CLS, Word Imbeddings