Machine Learning-Based Sparse Near-Field Scanning: When to Automatically Stop?

Active machine learning can significantly improve the efficiency of near-field scanning (NFS) by adaptively selecting important sparse samples. However, an appropriate and automated stopping criterion for the machine learning method has not yet been investigated and defined. This article proposes a novel automated stopping criterion for active-learning-based sparse NFS by estimating the worst-case prediction error at the unsampled locations. The machine-learning-based NFS process is automatically stopped when the estimated worst-case prediction error is below a user-defined threshold value, which can serve as a highly reliable and intuitive criterion to accommodate different requirements on the near-field image accuracy while maximizing the NFS efficiency. The proposed stopping criterion is validated in both simulation and measurement cases.