Model Robustness Under Distribution Shift

Measure covariate shift severity using Maximum Mean Discrepancy or KL divergence between training and deployment distributions

Each component has a specific meaning:

• Maximum Mean Discrepancy or KL divergence between training — The maximum mean discrepancy or kl divergence between training recorded for the patient or scenario being assessed.
• Deployment distributions — The deployment distributions recorded for the patient or scenario being assessed.

Note: Interpret the distribution shift result against the clinical thresholds and context described above.

Enter the Maximum Mean Discrepancy or KL divergence between training, deployment distributions for the patient or scenario you are assessing. Measure covariate shift severity using Maximum Mean Discrepancy or KL divergence between training and deployment distributions. Use the distribution shift result to inform your clinical assessment.