A Generalized Framework for Population Based Training
This work provides a more flexible and efficient hyperparameter optimization method for machine learning practitioners, though it is incremental as it builds on prior PBT approaches.
The authors tackled the problem of optimizing neural network weights and hyperparameters by proposing a general, black-box Population Based Training framework that supports asynchronous trials and dynamic schedules. They applied it to a WaveNet model for voice synthesis, achieving better accuracy, less sensitivity, and faster convergence compared to existing methods with the same computational resources.
Population Based Training (PBT) is a recent approach that jointly optimizes neural network weights and hyperparameters which periodically copies weights of the best performers and mutates hyperparameters during training. Previous PBT implementations have been synchronized glass-box systems. We propose a general, black-box PBT framework that distributes many asynchronous "trials" (a small number of training steps with warm-starting) across a cluster, coordinated by the PBT controller. The black-box design does not make assumptions on model architectures, loss functions or training procedures. Our system supports dynamic hyperparameter schedules to optimize both differentiable and non-differentiable metrics. We apply our system to train a state-of-the-art WaveNet generative model for human voice synthesis. We show that our PBT system achieves better accuracy, less sensitivity and faster convergence compared to existing methods, given the same computational resource.