Building trust in artificial intelligence: fusing interpretable machine learning and uncertainty quantification

We propose a method for interpretable machine learning (ML) with quantified uncertainty to address difficulty trusting ML by increasing transparency and robustness. The advantage over existing black-box ML methods (e.g., Gaussian process regression) is the proposed method produces inherently understandable model forms that can streamline the trust-building process. After completing this project NASA will have the capability to build ML models in a manner that better promotes trust building and a broadly accessible tool implementing this capability. We anticipate that the method may be better suited to some ML applications than others (i.e., regression analysis rather than classification); however, early work has already shown success in relevant fields such as physics and material science and this work aims to increase the diversity of applications tested.

The method will be illustrated on NASA application problem(s) (e.g., material modeling), compared to other methods, and packaged in accessible open-source software. The primary impacts of previous years of the project have been the research and development of the new ML method and the initial case-studies illustrating increased trust building potential. The target for transition is democratization of the technology: an open-source software package and illustration of success on relevant problems is the intended path to this transition. The risks of not selecting this project are (1) a lesser role of the center in NASA’s AI/ML development and (2) the loss of potential partnerships with Sandia NL and AFOSR on the topic.

This is early/mid-stage (TRL 4) — the most realistic path in is NASA SBIR/STTR, which funds small businesses and research institutions to develop technology aligned with NASA's needs (equity-free, phased funding). Check whether a current SBIR/STTR solicitation topic overlaps with this project's technology area, or contact the project directly (above) to ask.

None of these are guaranteed paths for this specific project — TechPort itself doesn't have an "apply" button. Reaching out to the contact(s) above with a specific question is usually the fastest way to find out what's actually open.