Faster Improvement Rate Population Based Training
This work addresses hyperparameter tuning automation for neural network training, offering an incremental improvement over PBT by enhancing long-term optimization.
The paper tackles the problem of Population Based Training (PBT) favoring short-term improvements over long-term performance by proposing Faster Improvement Rate PBT (FIRE PBT), which uses a novel fitness metric to focus on long-term gains, resulting in outperforming PBT on ImageNet and matching hand-tuned schedules while achieving faster learning and higher performance in reinforcement learning tasks.
The successful training of neural networks typically involves careful and time consuming hyperparameter tuning. Population Based Training (PBT) has recently been proposed to automate this process. PBT trains a population of neural networks concurrently, frequently mutating their hyperparameters throughout their training. However, the decision mechanisms of PBT are greedy and favour short-term improvements which can, in some cases, lead to poor long-term performance. This paper presents Faster Improvement Rate PBT (FIRE PBT) which addresses this problem. Our method is guided by an assumption: given two neural networks with similar performance and training with similar hyperparameters, the network showing the faster rate of improvement will lead to a better final performance. Using this, we derive a novel fitness metric and use it to make some of the population members focus on long-term performance. Our experiments show that FIRE PBT is able to outperform PBT on the ImageNet benchmark and match the performance of networks that were trained with a hand-tuned learning rate schedule. We apply FIRE PBT to reinforcement learning tasks and show that it leads to faster learning and higher final performance than both PBT and random hyperparameter search.