In these experiments, we used near field patches and used the loss function form the previous blog.
Summary:
| sigma of f | \(\lambda \) | learning rate | number of epochs | batch size | Q1 | Q2 | Q1 per bubble | Q2 per bubble | % gap between training and validation | |
| exp1 | 1 | 0.005 | 2e-5 | 300 | 1 | 13.450897 | 0.551683 | 0.527048 | 0.021537 | 30.9 |
| epx2 | 2 | 0.01 | 2e-5 | 300 | 1 | 21.001895 | 0.823452 | 1.218658 | 0.047734 | 8.8 |
Experiment 1: learning rate=2e-5, sigma of kernel f=1, regularizer parameter=0.005
Percentage gap between training and validation error ( \( \frac{validation errror – training error}{training error}*100 \) ) = 30.9
Average quantitative metric1 is 13.450897
Average metric1 per bubble is 0.527048
Average quantitative metric2 is 0.551683
Average metric2 per bubble is 0.021537
Worst Case:Center Location(x,z) = Average Metric1 per bubble: 0.902738, Average Metric2 per bubble: : 0.056981
Best Case:Center Location(x,z) = Average Metric1 per bubble: 0.470359, Average Metric2 per bubble: : 0.014894
Experiment 2: learning rate=2e-5, sigma of kernel f=2, regularizer parameter=0.01
Percentage gap between training and validation error ( \( \frac{validation errror – training error}{training error}*100 \) ) = 8.8
Average quantitative metric1 is 21.001895
Average metric1 per bubble is 0.823452
Average quantitative metric2 is 1.218658
Average metric2 per bubble is 0.047734
Worst Case:Center Location(x,z) = Average Metric1 per bubble: 0.919580, Average Metric2 per bubble: 0.060028
Best Case:Center Location(x,z) = Average Metric1 per bubble:0.789220 , Average Metric2 per bubble: 0.044518
