Model Validation via Precomputed Transformations (MVPT)

Model Validation via Precomputed Transformations (MVPT) is a proprietary approach to ensuring the reliability of Machine Learning (ML) models deployed in Autonomous Systems (AS) operating in extreme or data-limited environments. Traditional ML validation techniques fail to account for unknown or unmodeled scenarios, leading to potential mission failures when AS encounter novel conditions. MVPT addresses this by precomputing model validation across a set of anticipated data transformations before deployment, allowing real-time validation onboard the AS with minimal computational overhead. This enables autonomous decision-making without requiring human intervention or time-intensive model retraining. Phase I funding will be used to establish the feasibility of MVPT by conducting a comprehensive literature review of ML validation and transformation techniques, developing a working prototype, and defining a Phase II test plan for full-scale simulation in a NASA-relevant scenario. The target markets for MVPT include NASA’s space exploration and Earth observation missions, the Department of Defense (DoD) for autonomous Unmanned Aerial Systems (UAS), and private-sector industries such as autonomous vehicles, healthcare AI, and industrial robotics.