🤖 AI Summary
Population-Based Training (PBT)-style hyperparameter optimization (HPO) relies on manually specified hyperparameter update intervals—a critical meta-hyperparameter whose optimal value varies across tasks and lacks general design principles.
Method: We propose AutoPBT, a fully adaptive PBT framework that eliminates manual configuration of update intervals. Its core innovations include: (i) a task-agnostic weight reset mechanism; (ii) time-varying Bayesian optimization to dynamically schedule hyperparameter updates; and (iii) reuse of historical population weights to improve training efficiency.
Contribution/Results: AutoPBT removes the need for tuning this key meta-hyperparameter, enhancing both generality and adaptability. Evaluated on eight benchmark tasks—including image classification and reinforcement learning—AutoPBT consistently outperforms five state-of-the-art PBT variants and other HPO methods, achieving comparable or superior performance without incurring additional computational overhead.
📝 Abstract
Hyperparameter Optimization (HPO) can lift the burden of tuning hyperparameters (HPs) of neural networks. HPO algorithms from the Population Based Training (PBT) family are efficient thanks to dynamically adjusting HPs every few steps of the weight optimization. Recent results indicate that the number of steps between HP updates is an important meta-HP of all PBT variants that can substantially affect their performance. Yet, no method or intuition is available for efficiently setting its value. We introduce Iterated Population Based Training (IPBT), a novel PBT variant that automatically adjusts this HP via restarts that reuse weight information in a task-agnostic way and leverage time-varying Bayesian optimization to reinitialize HPs. Evaluation on 8 image classification and reinforcement learning tasks shows that, on average, our algorithm matches or outperforms 5 previous PBT variants and other HPO algorithms (random search, ASHA, SMAC3), without requiring a budget increase or any changes to its HPs. The source code is available at https://github.com/AwesomeLemon/IPBT.