Tag
This research paper investigates how data perturbations impact the accuracy and robustness of model cascades for efficient AI inference, identifying key failure modes and stressing the need for evaluation under distribution shift.