DPUQC: Data Pipeline Uncertainty and Quality Control for Trustworthy Space Autonomy

The DPUQC: Data Pipeline Uncertainty and Quality Control for Trustworthy Space Autonomy creates an updated pipeline that will treat Uncertainty Quantification (UQ) and Sensitivity analysis of models as primary objectives of the pipeline. This will be achieved by augmenting traditional pipelines, which currently handle problems of dynamic scaling, inhomogeneous data formats and data enrichment quite well. While these cloud native solutions have their merits, they do not consider the need to dynamically adjust operations in the presence of significant resource reduction. These pipelines assume that error analysis only the job of the terminal model. DPUQC will address this with a pipeline that pushes annotated data which carries quantifiers of uncertainty along with the data. This will provide data provenance for down stream decisions enabling explainability. The UQ that is transported can be adjust to accommodate changes in resources available to run the pipeline. Model analysis at the terminal edge of the pipeline will test model sensitivity to the variability of the data captured in the UQ, while also providing additional evidence to accept or reject the decision in the form of corroborating decisions from independently trained diagnostic models. The addition of provenance chains with UQ, diagnostic models and UQ model based testing will increase decision trustworthiness as each decision can be cross checked both via the model sensitivity and by comparing to additional evidence the diagnostic models provide. Any decision can be traced to the source that lead to it for diagnostic and quality control purposes.