{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/population-based-training-of-neural-networks","title":"Population Based Training of Neural Networks","arxiv_id":"1711.09846","date":"2017-11-27","proceeding":null,"authors":["Max Jaderberg","Valentin Dalibard","Simon Osindero","Wojciech M. Czarnecki","Jeff Donahue","Ali Razavi","Oriol Vinyals","Tim Green","Iain Dunning","Karen Simonyan","Chrisantha Fernando","Koray Kavukcuoglu"],"abstract":"Neural networks dominate the modern machine learning landscape, but their\ntraining and success still suffer from sensitivity to empirical choices of\nhyperparameters such as model architecture, loss function, and optimisation\nalgorithm. In this work we present \\emph{Population Based Training (PBT)}, a\nsimple asynchronous optimisation algorithm which effectively utilises a fixed\ncomputational budget to jointly optimise a population of models and their\nhyperparameters to maximise performance. Importantly, PBT discovers a schedule\nof hyperparameter settings rather than following the generally sub-optimal\nstrategy of trying to find a single fixed set to use for the whole course of\ntraining. With just a small modification to a typical distributed\nhyperparameter training framework, our method allows robust and reliable\ntraining of models. We demonstrate the effectiveness of PBT on deep\nreinforcement learning problems, showing faster wall-clock convergence and\nhigher final performance of agents by optimising over a suite of\nhyperparameters. In addition, we show the same method can be applied to\nsupervised learning for machine translation, where PBT is used to maximise the\nBLEU score directly, and also to training of Generative Adversarial Networks to\nmaximise the Inception score of generated images. In all cases PBT results in\nthe automatic discovery of hyperparameter schedules and model selection which\nresults in stable training and better final performance.","url_abs":"http://arxiv.org/abs/1711.09846v2","url_pdf":"http://arxiv.org/pdf/1711.09846v2.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"population-based-training-of-neural-networks","repo_url":"https://github.com/AlexHeyman/PopulationBasedTraining","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"unanswered"}},{"paper_slug":"population-based-training-of-neural-networks","repo_url":"https://github.com/JonasLeininger/ray-population-based-training","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"unanswered"}},{"paper_slug":"population-based-training-of-neural-networks","repo_url":"https://github.com/MattKleinsmith/pbt","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"unanswered"}},{"paper_slug":"population-based-training-of-neural-networks","repo_url":"https://github.com/cogsys-tuebingen/uninas","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"MIT"}},{"paper_slug":"population-based-training-of-neural-networks","repo_url":"https://github.com/facebookresearch/how-to-autorl","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"none","reach":{"status":"ok","spdx":"Apache-2.0"}},{"paper_slug":"population-based-training-of-neural-networks","repo_url":"https://github.com/ir2718/semantic-similarity-scoring","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"unanswered"}},{"paper_slug":"population-based-training-of-neural-networks","repo_url":"https://github.com/kklipski/ALHE-projekt","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"unanswered"}},{"paper_slug":"population-based-training-of-neural-networks","repo_url":"https://github.com/mle-infrastructure/mle-hyperopt","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"none","reach":{"status":"unanswered"}},{"paper_slug":"population-based-training-of-neural-networks","repo_url":"https://github.com/voiler/populationbasedtraining","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"unanswered"}}],"tasks":[{"task_slug":"deep-reinforcement-learning","task_name":"Deep Reinforcement Learning"},{"task_slug":"machine-translation","task_name":"Machine Translation"},{"task_slug":"model-selection","task_name":"Model Selection"},{"task_slug":"reinforcement-learning","task_name":"Reinforcement Learning"}],"methods":[{"method_slug":"population-based-training","method_name":"Population Based Training"}],"datasets_introduced":[],"methods_introduced":[{"slug":"population-based-training","name":"Population Based Training","full_name":"Population Based Training"}],"results":[],"syntology":{"syntology_url":"https://syntology.ai/paper/1711.09846","atlas_url":"https://app.syntology.ai/?focus=1711.09846","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}