Hyperparameter Optimization and Reproducibility in Deep Learning Model Training
This work addresses reproducibility issues for researchers in computational pathology, offering incremental guidelines based on empirical analysis.
The study tackled reproducibility challenges in foundation model training for histopathology by evaluating hyperparameter settings and augmentation strategies on a CLIP model trained with the QUILT-1M dataset, finding that specific RandomResizedCrop values (0.7-0.8) outperformed others and learning rates below 5.0e-5 consistently degraded performance across three downstream datasets.
Reproducibility remains a critical challenge in foundation model training for histopathology, often hindered by software randomness, hardware non-determinism, and inconsistent hyperparameter reporting. To investigate these issues, we trained a CLIP model on the QUILT-1M dataset and systematically evaluated the impact of different hyperparameter settings and augmentation strategies across three downstream histopathology datasets (PatchCamelyon, LC25000-Lung, and LC25000-Colon). Despite variability across runs, we identified clear trends: RandomResizedCrop values of 0.7-0.8 outperformed more aggressive (0.6) or conservative (0.9) settings, distributed training without local loss improved stability, and learning rates below 5.0e-5 consistently degraded performance across all datasets. The LC25000 (Colon) dataset consistently provided the most reproducible benchmark. These findings highlight that reproducibility in computational pathology depends not only on transparent documentation but also on carefully chosen experimental configurations, and we provide practical rules to guide future efforts in developing reproducible foundation models for digital pathology.