Diagnocat
Diagnocat: ML System Design
BatchNorm, mixed precision, ROC-AUC и кейс по 3D dental CT для instance segmentation маленьких periapical lesions при ограниченной разметке.
Аудио и материалы
Аудио собеседования
0:00 / 59:28
Этап 2 из 4DiagnocatML Engineer, медицинское CV2025-10-03 - 2025-10-22
Собеседования в Diagnocat: ML Engineer, медицинское CVML System Design в Diagnocat
Выводы и как готовиться
- The theory block is practical: not just definitions, but what statistics need to synchronize and what mixed precision changes in memory and numerics.
- The ROC-AUC discussion checks ranking intuition, score versus label inputs and when PR-AUC is more useful.
- The 3D case is the strongest ML System Design asset: crop around tooth instances, handle rare small lesions, choose segmentation/detection architecture and evaluate clinically useful instances.