Natural Language Explanation Generation on MIMIC-NLE modified (test)
67.34AUCWeNLEX
Evaluation Results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| WeNLEXTraining Mode=Weakly Supervised, Scenario=In-model (leaf symbol), Plausibility Loss (Plaus.)=MMD, Feature Reconstruction Loss (Recons.)=Yes, NLE Classification Loss (Clf.)=Yes, # NLEs=1115 (168)2026.03 | 67.34 | 79.91 | 10.61 | 95.65 | 0.3197 | 91.21 | 86.91 | 1.92 | 0.6704 | 45.38 | |
| WeNLEXTraining Mode=Weakly Supervised, Scenario=Post-hoc (snowflake symbol), Plausibility Loss (Plaus.)=Adv., Feature Reconstruction Loss (Recons.)=No, NLE Classification Loss (Clf.)=No, # NLEs=1038 (128)2026.03 | 65.13 | 82.47 | 4.876 | 94.72 | 0.2671 | 71.77 | 89.11 | 41.58 | 0.8079 | 36.21 | |
| WeNLEXTraining Mode=Weakly Supervised, Scenario=Post-hoc (snowflake symbol), Plausibility Loss (Plaus.)=MMD, Feature Reconstruction Loss (Recons.)=No, NLE Classification Loss (Clf.)=No, # NLEs=1038 (128)2026.03 | 65.13 | 82.47 | 7.601 | 91.95 | 0.3192 | 86.89 | 89.21 | 10.36 | 0.6525 | 50.39 | |
| WeNLEXTraining Mode=Weakly Supervised, Scenario=Post-hoc (snowflake symbol), Plausibility Loss (Plaus.)=MMD, Feature Reconstruction Loss (Recons.)=Yes, NLE Classification Loss (Clf.)=No, # NLEs=1038 (128)2026.03 | 65.13 | 82.47 | 9.303 | 92.48 | 0.3236 | 77.84 | 86.71 | 2.671 | 0.5258 | 47.5 | |
| WeNLEXTraining Mode=Weakly Supervised, Scenario=Post-hoc (snowflake symbol), Plausibility Loss (Plaus.)=MMD, Feature Reconstruction Loss (Recons.)=Yes, NLE Classification Loss (Clf.)=Yes, # NLEs=1038 (128)2026.03 | 65.13 | 82.47 | 9.401 | 91.48 | 0.3328 | 90.46 | 88.82 | 4.037 | 0.6286 | 47.35 | |
| Rio-Torto et al.Training Mode=Fully Supervised, Scenario=In-model (leaf symbol), Plausibility Loss (Plaus.)=MSE, Feature Reconstruction Loss (Recons.)=No, NLE Classification Loss (Clf.)=No, # NLEs=1107 (128)2026.03 | 64.08 | 83.29 | 10.37 | 97.63 | 0.565 | 89.97 | 88.26 | 8.785 | 0.7263 | 48 |