The False Promise of Zero-Shot Super-Resolution in Machine-Learned Operators

Research reveals that machine-learned operators (MLOs) fail at zero-shot super-resolution, unable to accurately perform inference at resolutions different from their training data. The study identifies key limitations in frequency extrapolation and resolution interpolation, proposing a multi-resolution training protocol as a solution.

• →Machine-learned operators cannot perform accurate zero-shot super-resolution inference on higher-resolution data than their training set.
• →MLOs fail at both frequency extrapolation and resolution interpolation tasks when tested in zero-shot scenarios.
• →The models are brittle and susceptible to aliasing when working with different resolutions than training data.
• →Researchers propose a computationally-efficient multi-resolution training protocol to address these limitations.
• →The findings challenge assumptions about MLOs' ability to handle arbitrary resolution inference in scientific computing.