A Generalized Framework for Population Based Training
By Ang Li, Ola Spyra, Sagi Perel, Valentin Dalibard, Max Jaderberg, Chenjie Gu, David M. Budden, Tim Harley, Pramod Gupta. 2019
73 citations recorded by the source corpus.
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 and faster convergence compared to existing methods, given the same computational resource.
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
- Bernd Bischl
- Anne-Laure Boulesteix
- Martin Binder
- Michel Lang
- Jack Parker-Holder
- Aldo Pacchiano
- Tobias Pielok
- Jakob Richter
Citations in the corpus
- Hyper-Parameter Optimization: A Review of Algorithms and Applications (2020, methodology): “PBT is quite a unique algorithm in its combination of the process of hyperparameter tuning and model training.”
- A System for Massively Parallel Hyperparameter Tuning (2018, methodology): “Established parallel methods for hyper-parameter tuning include PBT (Jaderberg et al., 2017; Li et al., 2019) and Vizier (Golovin et al., 2017).”
- Hyperparameter optimization: Foundations, algorithms, best practices, and open challenges (2021, background): “This can be reduced by combining PBT with ideas from bandits and BO (Parker-Holder et al., 2020).”
- Effective Diversity in Population-Based Reinforcement Learning (2020, background): “Training a population of agents provides is a promising approach to gathering a diverse range of experiences, often with the same (wall-clock) training time (Recht et al., 2011; Li et al., 2019).”