Top: noise-corrected normalized correlations between giant pulse data and predicted IRFs for 0 ms ≤ τ ≤ 0.8 ms, where most of the signal lies. The correlations are corrected for contributions by noise in both the data and model but not for correlation in the noise between the two. Bottom: normalized correlations for the pure-noise regions of τ < 0 and >3 ms between the data and model. The correlation between the model and the training set confirms that they share correlated noise, while the lack of correlation with the validation set shows that there are no systematic effects shared between all pulses.