KL Divergence Comparison for CheXpert (averaged over 1 runs):
+---------------------------+---------------------+-----------------------+
| Method                    | Average KL (Prob)   | Average KL (Argmax)   |
+===========================+=====================+=======================+
| Original                  | 4.38e-02 ± 0.00e+00 | 1.70e-01 ± 0.00e+00   |
+---------------------------+---------------------+-----------------------+
| Arch Mod                  | 5.17e-02 ± 0.00e+00 | 1.54e-01 ± 0.00e+00   |
+---------------------------+---------------------+-----------------------+
| Replace Mean              | 2.89e-02 ± 0.00e+00 | 8.57e-02 ± 0.00e+00   |
+---------------------------+---------------------+-----------------------+
| PatchCutout-trained Model | 5.34e-03 ± 0.00e+00 | 2.92e-02 ± 0.00e+00   |
+---------------------------+---------------------+-----------------------+
| Temperature Scaling       | 9.80e-03 ± 0.00e+00 | 1.71e-01 ± 0.00e+00   |
+---------------------------+---------------------+-----------------------+
| Platt Scaling             | 1.46e-01 ± 0.00e+00 | 1.75e-01 ± 0.00e+00   |
+---------------------------+---------------------+-----------------------+
| MCal_CE (Cross-Entropy)   | 8.01e-06 ± 0.00e+00 | 3.77e-03 ± 0.00e+00   |
+---------------------------+---------------------+-----------------------+