The Model Analyst's Toolkit: Scientific Model Development, Analysis, and Validation
Abstract
One of the most powerful things scientists can do is to create models that describe the world around us. Models help scientists organize their theories and suggest additional experiments to run. The proposed research effort builds on and extends the work of the previous ONR-funded Validation Coverage Toolkit for HSCB Models project, [ONR contract N0014-09-C-0463]. The overall objectives of the ongoing research program are: (1) Help scientists create, analyze, refine, and validate rich scientific models. (2) Help computational scientists verify the correctness of their implementations of those models. (3) Help users of scientific models, including decision makers within the US Navy, to use those models correctly and with confidence (4) Use a combination of human-driven data visualization and analysis, automated data analysis, and machine learning to leverage human expertise in model building with automated analyses of complex models against large datasets. Specific objectives for the current effort include: (A) Fluid temporal correlation analysis. (B) Automated suggestions for model construction and refinement. (C) Data validation and repair. (D) System prototyping. (E) Evaluation of applicability to multiple scientific domains. During the current reporting period, we focused primarily on improving the causal analysis functionality of MAT. This included three subtasks: using the MAT tools for a real-world analysis to demonstrate and evaluate the various causal analysis methods, adding a new causal analysis method that we hope will provide better performance for certain kinds of analysis in the face of significant noise in the data, and create a centralized method for running and summarizing the results of the various analysis methods.
Document Details
- Document Type
- Technical Report
- Publication Date
- May 20, 2015
- Accession Number
- ADA619964
Entities
People
- Scott N. Reilly