Efficient Population Based Hyperparameter Scheduling for Medical Image Segmentation

Abstract

The training hyperparameters (learning rate, augmentation policies, e.t.c) are key factors affecting the performance of deep networks for medical image segmentation. Manual or automatic hyperparameter optimization (HPO) is used to improve the performance. However, manual tuning is infeasible for a large number of parameters, and existing automatic HPO methods like Bayesian optimization are extremely time consuming. Moreover, they can only find a fixed set of hyperparameters. Population based training (PBT) has shown its ability to find dynamic hyperparameters and has fast search speed by using parallel training processes. However, it is still expensive for large 3D medical image datasets with limited GPUs, and the performance lower bound is unknown. In this paper, we focus on improving the network performance using hyperparameter scheduling via PBT with limited computation cost. The core idea is to train the network with a default setting from prior knowledge, and finetune using PBT based hyperparameter scheduling. Our method can achieve 1%–3% performance improvements over default setting while only taking 3%–10% computation cost of training from scratch using PBT.