GRASP: Global Reading and Assembly for Semantic, Probabilistic World Models

Abstract

Reading: - Added additional preprocessing step to clean the input to the extraction framework. - Various modifications to further optimize the Eidos runtime - Running the VariableFinder from IHMC to align ontology nodes with DSSAT variables - Established plan for unifying upper-level ontology between BBN and Eidos, potentially with Sofia too - Collaborated with BBN to develop strategy for incorporating/demoing BBNÕs LearnIt tool Assembly Framework (HMS): - Developed new API and processor for Sofia web service and JSON format in INDRA - Integrated Eidos and Sofia text reading into INDRA REST API - Developed Bayesian belief scoring framework to support real-time feedback and belief adjustments Probabilistic Assembly: - Integrated DSSAT data on precipitation vs crop yields for the lean season in South Sudan for Maize and Sorghum in the states of Unity and Northern Bahr el Ghazal. - Implemented parameterization using a database as a source instead of CSV files, speeding it up considerably. - Implemented new algorithm that ensures propagation of causal influences beyond one hop, updated the model writeup here (section 1.4 newly added): http://vision.cs.arizona.edu/adarsh/export/Arizona_Text_to_Model_Procedure.pdf Context: - Completed branch of timenorm system that further speeds things up by reducing the context window size - Refined geonorm system and submitted to SemEval 2019: Task 12: Toponym Resolution in Scientific Papers, a shared task for evaluating geographical name normalization systems

Document Details

Document Type
DoD Grant Award
Publication Date
Feb 14, 2019
Source ID
W911NF1810014

Entities

People

  • Mihai Surdeanu

Organizations

  • Army Contracting Command
  • Defense Advanced Research Projects Agency
  • University of Arizona

Tags

Readers

  • Aquatic Ecology
  • Artificial Intelligence
  • Database Systems and Applications

Technology Areas

  • AI & ML
  • AI & ML - Bayesian Inference
  • AI & ML - Information Retrieval